AD Ai38 151 R STUDY OF OPTIMAL COMPUTER NET14ORK ARCHITECTURE FOR 1/3
WRGTPTESNAE H HUDIGITAL AVIONICS 5 5 EMSU H FI FORCE INS OF TECH
UNCLASSIFIED DEC 83 AFIT/GE/FEE?-1-6 F/G1/2 N
EEEEEmmhEmhlhhIEhEmhEEohmhmhEEohhohEEEmhhEIEhhmhEmhEEEEEEEhhEEEEEEEEEEEEhohEEEEEmhEEE
11.2.
2.0.
Nmo
L%11111.25 ~ll i4 1.
*qirP.
U.. %*% No V-
OF
A STUDY OF OPTIMAL COMPUTER NETWORK*'. ARCHITECTURE FOR DIGITAL AVIONICS
SYSTEMS
THESIS
Mensur F. Krilic, Lt ColYugoslavian Air Force
AFIT/GE/EE/83D-36
Cm
S-rE
L"..J "hio Ind re,. it 2 3 1984[=" tc ~~L publi: Ie],?.J. ,. 'd =,h.w .
LLdi t:ibution is utiil.,ited.
DEPARTMENT OF THE AIR FORCE
E..AIR UNIVERSITY
AIR FORCE INSTITUTE OF TECHNOLOGY
Wright-Patterson Air Force Base, Ohio
4.€
"h'S. ."L.". ,."". '- '. . " ."."-"'".-i-.'- ' '-'.-' " - "" " , - ', . """" ."-"""'"""""'. . """ " "
AFIT/GE/EE/8aD-36."q
"I
A STUDY OF OPTIMAL COMPUTER NETWORKARCHITECTURE FOR DIGITAL AVIONICS
SVSTEMS
THESIS
Mensur F. Krilic, Lt ColYugoslavian Air Force
AFIT/GE/EE/83D-36
- f2TE
S2 3 1984
e fr pA
Approved for public release; distribution unlimited
AFIT/GE/EE/83D-36
A STUDY OF OPTIMAL COMPUTER NETWORK ARCHITECTURE
FOR DIGITAL AVIONICS SYSTEMS
THESIS
9,
Presented to the Faculty of the School of Engineering
of the Air Force Institute of Technology
Air University
in Partial Fulfillment of the
. Requirements for the Degree of
Master of Science in Electrical Engineering
T T -
<: .....-f-.
Mensur F. Krilic, B.S.
Lieutenant Colonel, Yugoslavian Air Force----
t .December 1983
. "--' Approved for public release; distribution unlimited
4.2
.4. r '-I ,
Preface
The purpose of this thesis was to define the characte-
- ristics of optimal computer network architecture for digital
avionics systems. Previous investigations in the field have
indicated that such a definition should be made today to
meet the challenging requirements of future avionics.
Although the scope of this work was limited by availab-
4le unclassified data about military avionics systems, I
believe that this report will be found complete and self-
sufficient by the reader interested in the subject.
I wish to thank my advisor, Major W.D. Seward who pro-
.1 posed this project and to the thesis committee members LtCol
0 H.W. Carter and Capt J.S. Gordon of the Air Force Institute
of Technology. I greatly appreciate their willingness to
* help and confidence to let me work independently when I
needed it.
9A special gratitude is extended to Mr. S.J. Larimer of
the Flight Dynamics Laboratory, WPAFB for his comments and
readiness to review the design of Self-Managing Multiplex
System.
S And finally, I would wish to thank my wife, Bahra, and
our daughters, Amela and Alma, for their generous support
and understanding during these very busy months.
Mensur F. Krilic
hp ii'€.. : . ' t .,.¢.i .. [ . " k '"'/..".'( ' t:;: :
Table of Contents
Page
Preface ............ 4.................. ............. ii
.4List of Figures ............................... *...... vi
List of Abbreviations ........................... **4 X
Abstract .. . . . . . . . . . . . . . . . . . . . . . xi
I. Introduction ............ **... . . . ..... **44 . ..* 1
Ii. Results of Literature Search .................. 7
III. Characteristics of Avionic's Architectures .... 14
Multiplex System "Descriptive Issues" ....... 16
System Architecture ..... *............. 16B us Control ...............................44 16System Topology ... 4............4........ 16Message Protocol ........... ............... ~ 18Transmission Medium ....................... 18Coupling Method ... . . . ... . .. . .*. . .. . ... .. . 20Signal Modulation 4............ ........... 25Synchronization . ....... .. ...... ........*~... 32
Multiplex System "Analysis Issues" ... 0 .. 34
Bus Allocation Scheme ..................... 35Bus Efficiency and Utilization ............ 41Data Throughput and Bus Bandwidth .. *..... 41Message Latency and Overhead ........ 43Word Length ............................... 43Message Addressing ........................ 45Periodic versus Aperiodic Messages ... 444.4 47Block Messages Transfer ................... 49System Integrity 0...... .0 ....... 6-0 50Functional Partitioning ........... 0.......* 51Flexibility ............................... 52Software Interface ........................ 53
Chapter Summary . .. . .. .. .. .. . . ..............4 4 . 57
Page
IV. Avionics Architectures Descriptionand Analysis . .. .. .. . ... . ... .. .. .. .. .. .. .. . . ... 59
The F-15 Multiplex System ................... 60The B-i Multiplex System .................... 66
The F-16 Multiplex System ................... 72DAIS - Digital Avionics Information System .. 78
*DAIS Processors .................*......... 80DAIS Multiplex Bus ... .... ......... ...... .. 80DAIS Controls and Displays ................ 81Mission Software .......................... 81
GAMS - General Aircraft Multiplex System .... 83
Terminal ...... *......................... 85Multiplex Cable ........................... 85Controller .. . ... ... .. . . **. . *.................** 86Monitor .... **............................ 86
HXDP - Honeywell ExperimentalDistributed Multiplex System ................ 91Loop Multiplex System .............. 0......... 97Passive Bus System . .. .. .. .. .. . .. ... . ... .. . .. 103Active Bus System ................ ... .. .. .... 108Switched Network ............................ 11lDP/M Distributed Processor/Memory System .... 117Transparent Contention System .... ... ... .. .. . 122
Transparent Contention Concept ............ 124Virtual Memory Technique .... ....... 127Software Partitioning ..................... 130Task Volunteering . ... . .. .... . .. .. . ... . .. .. 132
Frequency Multiplex System .................. 1354 ARINC-575 Digital Air Data Multiplex
System .......... ......................... 140Chapter Summary ......... ~*......*...*...... 141
V. Optimal Architecture Definitionand Conceptual Design ..... e.............*..e... 143
Comparison with the AE-9B "Future" Bus *... 146Self-Managing Multiplex System .. ..... ... .... 149
JBasic System Requirements ................. 149Transmission Medium . ... .. .. . .. .. .. . .. . .... 150Signal Modulation .. ... . .. . .. ... .. .. . . .. ... 151Bus Allocation Scheme ..................... 151
iv
Page
Message Protocol ..... *..................... 156BuslInterface Unit ........................ 163Acknowledgments in Broadcast Mode ......... 169System Network Topology ................... 174Redundancy in the System .................. 176Bus Leng th . . . .. **....e*.*e.. 180
Number of Terminals ........... 182Block and Trigger Messages Transfer ....... 185B us Efficiency . .. . ... .... *........ . ... .. . .. 186Bus Utilization and Message Latency ....... 187Software Interface ........................ 189
VI. Conclusions and Recommendations ............... 192
Bi l o r p y . . . .. . . . . . . . . .. . . . . . . 9
4..
-----.
List of Figures
Figure Page
1 Bus Topologies .............................. 17
2 "Lossless" Line Configuration ............... 21
3 Looped Lossy Line Configuration ............. 23
4 Looped Lossy Cross-Coupled Configuration .... 23
5 Comparison of Lossless andLossy Configurations ........................ 25
6 Unipolar and Polar NRZ Modulations .......... 27
7 Polar RZ Modulation ......................... 27
8 Bipolar NRZ Modulation ...................... 28
9 Delay Modulation ............................ 29
10 Duobinary Modulation ........................ 29
11 BiPhase-Level Modulation .................... 30
12 Example of Hierarchical Message Addressing .. 46
13 Major and Minor Cycles ...................... 48
14 Typical Example of Life Cycle Curves ......... 55
15 The F-15 Data Bus Architecture .............. 61
16 The F-15 Multiplex System Wordand Message Formats ...................... .... 64
17 The B-1 AMUX Bus Architecture ............... 68
18 The B-1 Multiplexed SerialData Word Format ............................ 70
19 The F-16 Multiplex System Architecture ...... 73
20 The 3-bit Illegal Manchester-IISync for Command and Status Words ........... 74
' vi
5' T , _. . ,. = .% . ,. . • ,. ;. .
9 - < . . '. ". - , '-, "- ' . "'e '. " .""."" '., ""' - """ ""-, . "' :-
Figure Page
21 The 3-bit Illegal Manchester-IISync for Data Words .. .. .. ........... .. . .. 74
22 The F-16's Word Formats ..................... 76
23 DAIS Architecture - Hardware ................ 80
24 DAIS Architecture - Software ................ 82
25 GAMS Architecture ........................... 84
26 GAMS Message Formats ........................ 87
27 HXDP System Configuration ................... 91
28 HXDP Message Format ......................... 93
29 Vector-Driven ProportionalAccess Mechanism . ... . . ... .. . ... ... .. .. . .. . .. 94
30 Loop System Topology *..................... 98
31 Pierce Loop Organization .................... 99
32 Loop Word Format ..... ... .... .. ..... ... ... .. . 100
33 Packet Format for Loop System ............... 10i
34 Passive Bus System Topology ................. 104
35 Active Bus Topology ......................... 109
36 Switching Network Architecture .............. 113
-37 Switched Network Message Types .............. 114
38 Switched Network Command Table Format ....... 115
39 Texas Instruments DP/M Configuration ........ 118
40 Honeywell DP/M Hybrid Bus Topology .......... 119
41 Honeywell DP/M Switched Local Buses ......... 120
42 Transparent Contention Conceptual Topology .. 123
-. 43 Transmitter-Bus Interface ................... 124
* vii
Figure Page
44 Bus Contention Arbitration .................. 125
45 Transmitter Circuit ......................... 126
46 Implementation of Virtual Memory ............ 128
47 Receiver Block Diagram ...................... 129
48 Directional Frequency Multiplex Bus -Position Dependent Configuration ............ 136
49 Directional Frequency Multiplex Bus -Head End Configuration ...................... 137
50 Two Cable Frequency Multiplex Bus ........... 139
51 The SMS Transparent ContentionImplemented by Manchester-II Coding ......... 153
52 The SMS Message Protocol .................... 157
53 The SMS "Book-Keeping" Word ................. 160
54 The SMS Bus Interface Unit BlockDiagram ........................... ........ 164
55 The SMS Transmitter-Receiver Unit ........... 165
56 The SMS Network Topology .................... 175
57 The SMS Inter-Bus Interface Unit ............ 176
, viii
List of Tables
Table Page
I Typical Average MaintenanceWorkloading ................................. 54
II Prime Characteristics Required forFuture High-Speed Data Bus .................. 147
i "x
...1g
*te
-A; List of Abbreviations
' AAES - Advanced Aircraft Electrical SystemACU - Avionics Control UnitAMUX - Avionics Multiplex SystemARPV - Advanced Remotely Piloted VehicleBITE - Built-In-TestBIU - Bus Interface UnitBCIU - Bus Control Interface Unitbps - Bits per SecondCRC - Cyclic Redundancy CheckCTIS - Centrally Integrated Control SystemDAIS - Digital Avionics Information SystemEMI - Electromagnetic InterferenceEMUX - Electrical Multiplex SystemEOT - End of TransmissionFAA - Federal Aviation AdministrationFIFO - First-In-First-OutFCC - Fire Control ComputerGAMS - General Aircraft Multiplex SystemGNACU - General/Navigation Avionics Control UnitHSMB - High-Speed Multiplex BusHXDP - Honeywell Experimental Distributed ProcessorITS - Information Transfer SystemLIU - Loop Interface UnitMC - Mission ComputerMTBF - Mean Time Between FailuresNADC - Naval Air Development CenterNRZ - Non-Return-To-ZeroPE - Processing ElementRT - Remote TerminalRZ - Return-To-ZeroSIM - State Information MemorySOT - Start of TransmissionSMS - Self-Managing Multiplex SystemTDM - Time Division MultiplexTSP - Twisted-Shielded PairUSAF - United States Air ForceWDACU - Weapon Delivery Avionics Control UnitWDM - Wavelength Division MultiplexingWPAFB - Wright-Patterson Air Force Base
x
S7 - a 7 P.
'.,.. AFIT/GE/EE/83D-36
Abstract
The ramifications of the use of digital avionic systems
and the experience drawn from the 10-year development cycle
of the MIL-STD-1553B require that solving of the problems
- - due to anticipated growth of avionics should start today.
Emerging technologies offer more and more capabilities to
avionic systems and optimization of computer architectures
is needed to obtain all the benefits that potentially exist.
In this paper, current theoretical considerations in
available literature have been used to sort out the essen-
4' tial figures of merit of computer network architectures for
digital avionic systems. Fourteen different approaches to
the same problem of data multiplexing in avionics systems
are analyzed according to the key issues. Conclusions drawn
are used to define the "optimal" computer network architec-
ture for digital avionics.
The Self-Managing Multiplex System (SMS) is conceptu-
ally designed with respect to the "optimal" characteristics,
along with the discussions of some trade-offs that had to be
made. The burst errors self-correcting feature of the broad-
cast-acknowledgments in the SMS concept seems to deserve
some sort of testing in practice. It is recommended that a
detailed simulation study should be performed later and a
hot bench built up using the latest technologies that exist.
'a xi
,
-. .w'*. * -*1> - - -7. 7 -777. 7
:~ ~ A STUDY OF OPTIMAL COMPUTER NETWORK ARCHITECTUREFOR DIGITAL AVIONIC SYSTEMS
I. Introduction
A At a first glance, the transmission of multiplexed data
over a pair of simple twisted-shielded wires might be con-
sidered as a problem that is not broad enough to warrant the
thesis leading to the Master of Science Degree in Electrical
Engineering. However, the subject is far reaching indeed,
simply due to the fact that the pair of wires represent the
internal time division multiplex data bus -- the "artery" of
any digital avionic system. All impacts of data multiplexing
can be understood only when the ramifications of the use of
P digital avionic systems are considered: since the first
digital computer as has been employed in the F-106 in the
late 1950's, up to the so-called "all-digital aircraft" as
foreseen for the 19901s.
Digital avionics multiplex techniques have a number of
definite advantages over their analog counterparts. Elimina-
tion of unnecessary replication of information sensing and
display, performance and reliability gains, reduction of the
weight, size and costs, and lack of space are usually given
as the major reasons for the digital avionics usage and
integration (Ref 3:2-2). -As pointed out in the same refer-
ence, digital avionics integration is the cooperative use of
%' *
t-.....~ shared information among avionic subsystems, which became
necessary when requirements for missions and their associa-
ted avionic hardware could no longer be met practically in
an aircraft with independent and self-sufficient subsystems.
On the other hand, the classical avionic systems design
methodology was to provide unique systems for the specific
aircraft. Different manufacturers have produced the hardware
with little or no relationship to each other. The prolifera-
tion 3f "black-boxes" in the aircraft that were part of the
same fleet, but without any compatibility or commonality of
either hardware or software for their avionics, resulted in
the excessive overall avionics design, acquisition, and
maintenance costs that tended to restrict the number of
aircraft procured. In that sense, not only integration with-
in a single aircraft was needed, but the integration had to
*be done in a way that provides standardization among the
avionic systems even when they were intended for completely
different missions. The time division multiplex data bus is,
undoubtedly, the cornerstone of both the digital avionics
integration and standardization.
The standardization efforts started in the late 1960's-
early 1970's and coincide with the F-15's, B-i's, and F-16's
"* development or early procurement stages. As the first result,
the USAF issued its MIL-STD-1553 standard in August 1973.
Based on the application requirements for the USAF, US Navy,
* 2
4 * , .*w
* * ~ . * ~ S - . P 2 .. .. . . . -
• -and US Army, the standard has been upgraded and published as
the tri-service document, the MIL-STD-1553A on 30 April
1975. The current update, MIL-STD-1553B was issued on
a21 September 1978 with the Notice 1 (USAF) from 21 February
1980. The standard is accepted by Great Britain, as well.
The technology is considered mature today. The MIL-STD-
1553A/B-based hardware components, software solutions and/or
entire system architectures can be found almost everywhere
throughout the aerospace industry. More and more weapon sys-
tems are built around the MIL-STD-1553 bus today. According
to the latest published data (Ref 31), this is the case
even when severe environmental, space, weight, and power
a restrictions limited the amount of computing capability
available to support bus operation. An example is the MSRAM
(medium range air-to-surface missile) which is an adapta-
tion of the Tomahawk cruise missile.
Digital avionics technology continues to increase in
number of applications and complexity by an order of magni-
tude every few years (Ref 25:48). Today, avionics account
for about one-third of the aircraft's total cost. In paral-
lel, the MIL-STD-1553 philosophy has proven itself to be an
effective way for avionics integration and will, very proba-
bly, stay that way well into the 1990's. The question then
arises: 'What is the purpose of further research to change
an effective standard that fits so well to such complex and
a- 3-a
" " '''€ €''g a * . € "".'.. ."%""-V .r" '""- ' " "'€ ' ''-"".''."'- - '
., j d , "% ' "' V ." "'" "~* "'' * " .""'.-'".*- " "- "~ .' -. "" ""
expensive equipment?'
The twofold answer to that question covers the scope of
this thesis, as well:
1. The anticipated growth of avionics will require more
and more processing capabilities and, therefore, more and
more data to be transferred at high rates. Having in mind the
10-years development cycle of the MIL-STD-1553B, it is obvi-
ous that solving problems of tomorrow should start today. In
addition, some inherent inflexibility of the MIL-STD-1553
technique, which is hidden by the undisputable advantages of
the approach, might cause a lag in the ability to exploit
all benefits from the emerging technologies in the future.
2. The exact network topology, data protocol, and con-
figuration have not been standardized for the avionic subsys-
tems themselves, although they must be designed around the
MIL-STD-1553 data bus. An analysis is required to determine
the best subsystem network architecture for optimal system
response and reliability.
In fact, this is a subject which has received much at-
tention in recent years. Nearly every organization involved
in aerospace system design has its own research program
in this area (Ref 53:309). A huge amount of literature
* exists, with sometimes diametrically opposite opinions ex-
pressed. The survey of literature search, as given in Chap-
ter II of this thesis, will help the interested reader
O-P %
u4
54* locate valuable information to start his own research. In
addition, throughout the report numerous citations from a
variety of references are made to provide the reader more
insight into the subject area.
The discussion in Chapter III is based on existing
~1 general theoretical considerations about the figures of mer-
it of different philosophies. The key issues are sorted out
in an attempt to help description and analysis of a variety
of approaches to the same problem of data multiplexing.
In Chapter IV, fourteen different techniques are de-
scribed and analyzed with respect to the key issues dis-
cussed in Chapter III.
The conclusions drawn from both Chapter III and Chapter
IV led to a definition of what the "optimal" architecture
should be. The characteristics of the architecture are list-
ed in Chapter V. In order to resolve some contradictions in
this set of characteristics of an "optimal" architecture,
.4 and to give guidance how to synthesize a system by resolving
the contradictions on a problem-by-problem basis, the Self-
Managing System (SMS) has been introduced in the same chap-
ter. Except for the concept that allows acknowledgments in
broadcast mode, the technique is not a completely new idea.
Rather, it is an integration of several different techniques
* in a new context.
5
The report concludes in Chapter VI. A limit on the
:1: study was the security clearance required to obtain complete
information about existing solutions on military aircraft,
including their weaknesses and strength. The author had no
such a permission and the results are based on the available
data from unclassified literature. This means that the as-
sumptions made might not be true in filed applications. To
truly prove a suitability of the SMS technique, a recommnen-
dation is made in Chapter VI that a detailed simulation
study should be performed later and a hot bench built up
using the latest existing technologies. At least, the burst
errors self-correcting feature of the broadcast-acknowledg-
ments concept seems to deserve some sort of testing in prac-
tice.
*16
K4U. V-- - .6.-
~ -~ II. Results of Literature Search
In order to identify and evaluate sources of informa-
tion, two separate literature searches were conducted:
through the National Aeronautics and Space Administration
(NASA Literature Search Number N49276 / May 1983), and the
Defense Technical Information Center (DTIC / April 1983). In
addition to that, the author has performed an independent
search over available literature at the Academic Libraries
at the AFIT School of Engineering and School of Systems and
Logistics, the AFWAL/TST Technical Library in WPAFB, and the
Wright State University Library in Fairborn, Ohio.
These searches include accessible unclassified journals,
reports, and technical information published since 1968 on
4'4' both the existing and proposed solutions pertaining to digi-
tal avionics integration, multiplexing, and architectures.
In that way, the period from early days of digital avionics
and the first attempts for standardization in the field is
covered.
More than five hundred references were reported in
these searches, showing continued efforts made over the past
15 years to find an optimal solution in the area. The high-
lights of the literature, as a combined list of references
and bibliography, can be found at the end of this report.
In spite of the fact that there are many apparent
7
* controversies involving both newly proposed and past model-
ing techniques, it is noticeable that avionic architectures
.,* S.are generally evolving towards more sophisticated capabili-
ties offered by hierarchical, distributed control architec-
tures.
On the other hand, the main characteristic of present
4 solutions is an architecture based on the MIL-STD-1553 Air-
craft Internal Time Division Command/Response Multiplex Data
Bus (Ref 65). Although there is not a reference without some
sort of description, critique, or eulogy of this solution,
the most complete survey of the bus can be found in the
Multiplex Application Handbook (Ref 3), with shorter ver-
sions offered by Ref 21 and 48. Useful literature about this
topic is also represented by Ref 36, 37, 62, 73, and 75,
while a detailed analysis of some of the main physical
characteristics of the bus is given in Ref 57. A comparison
$ based on a computer simulation of three somewhat different
implementation of the MIL-STD-1553 based data buses is of-
fered in Ref 58. Many solutions can be found in literature
for bus controllers, remote terminals, different types of
testers, and simulation design tools pertaining to the MIL-
STD-1553 architecture (Ref 5, 14, 17, 29, 33, 44, 47, 60,
64, 78, 84, and 91).
Having in mind that incompatibilities exist between
military and commercial aircraft data multiplexing systems,
8
.5.*
*~'. SCI Systems Inc. developed the Programmable Interface Module
(Ref 41) which could be used in achieving the compatibility
* between different buses.
Air superiority, close air support, manned bomber, and
remotely piloted vehicles are recognized in Ref 20 as having
a large core of common equipment types performing fairly
similar functions across the various aircraft. This fact led
to the definition of the general system framework known as
the Digital Avionics Information System (DAIS), being under
* development by the Air Force since 1973. DAIS provides a
system architecture which can adopt currently available
avionics subsystems (radars, inertial platforms, etc.) to a
common, modularly expandable avionics core (Ref 87), repre-
0 senting the second most often referenced concept throughout
the literature (Ref 8, 9, 10, 11, 12, 16, 23, 32, 59).
Results of an investigation of functional commonality
of avionics systems specific to Naval aircraft, with general
requirements for the development of core Naval aircraft
avionics equipment are given in Ref 22.
An attempt for computer system optimization for avion-
ics application is described in Ref 15. This was made by an
examination of the commonalities among the four subareas of
signal processing (radar, communications, image processing
and electronic warfare), in order to determine computational
'4 complexity and commonality among kernels and algorithms for
* .. **~'0~'q *r ~ 99 ' ' 9 , .
* .*.,architectural purposes.
Conceptual design considerations for an integrated
power and avionics information system can be found in Ref
18, 28, and 38.
Almost immediately after the adoption of the firstversion of the MIL-STD-1553 architecture, some disadvantages
of the concept have been recognized (Ref 80) and the search-
es for other solutions are still continuing (Ref 19). The
main goal is to achieve an architectural concept that will
allow multimission applications along with increased data
rates and improved reliability, survivability, availability
and affordability. This has been analyzed and presented in
an excellent study made by the Boeing Military Airplane
Company (Ref 81) in late 1982.
The McDonnell Aircraft Company is currently performing
further studies of an advanced avionics system specification
within the Pave Pilar Advanced System Integration Demonstra-
-. tion (ASID) Definition Program. This study is contracted
(F33615-82-C-1904) by the Avionics Laboratory Division of
the Air Force Wright Aeronautical Laboratories (AFWAL) and
will be done in 1984; the initial results were presented at
*! the NAECON '83 (Ref 76).
In fact, nearly every military or civilian institution
dealing with design of airborne systems has its own research
program in this field (and its own opinion and approach to
10
• '. the problem!), as follows:
In the Avionic Multiplex System (AMUX) developed by the
Vought Corporation for the Advanced Aircraft Electrical.
System (AAES) Simulator at the Naval Air Development Center
(NADC), both the command/response and polled contention
modes are implemented (Ref 90), capitalizing in this way
upon the dynamic bus allocation feature of the MIL-STD-1553
concept.
But, this is not an isolated attempt to fully explore
all of the existing features of present standard. In Ref 4,
the British Aerospace PLC addresses problems with regard to
the handling of acyclic message transfers either on the same
bus or between buses, and to the broadcast transfer checking
schemes. The High Speed Multiplex Bus (HSMB) Protocol Study,
made by NADC (Ref 86), makes not only the use of MIL-STD-
1553B functions as a subset of an HSMB protocol, but also
goes beyond to completely specify the broadcast mode of
operation and a more detailed implementation method for
dynamic bus control allocation. A so-called Hybrid System is
recommended by Texas Instruments Inc. (Ref 2) in an attempt
to design an avionic digital processing system for multimis-
sion Advanced Remotely Piloted Vehicle (ARPV) applications.
,A bus-oriented modular avionic architecture with dis-
. tributed processing seems to offer highly desirable attri-
butes of growth and independence from technology implementa-
11
* .-. '> ." **~ **.. *.PL.. .
* tion details. This is the philosophy advocated by General
Electric in Ref 67, using the US Navy's Advanced Integrated
Display System as the example. It must be noted that this
philosophy is, in a way, the ultimate goal being implicitly
or explicitly advocated by almost everybody. Controversies
arise concerning different opinions as to what is the "best"
way to achieve that goal.
In Ref 63, Boeing Military Airplane Company pleads for
contention multiplex systems and therefore for the subsystem
4 independence that was lost by integrating with a central
minicomputer. The quantitative performance of three various
types of contention Information Transfer Systems (ITS) is
presented and the conditions for which particular ITS would
0 be appropriate are discussed.
Ref 53, 55, and 61 are showing the efforts made by the
AFWAL to achieve an architecture for event-driven real-time
systems, which is diametrically opposite to the MIL-STD-1553
* concept and represents an attempt to explore all potential
benefits of distributed architecture and purely (so-called
"transparent") contention protocol.
General Dynamics proposed a switched network archi-
tecture (Ref 50) that works very much like the public tele-
.4 phone system, dropping out the bus oriented technique, but
still providing a support for modular structures.
* In an excellent tutorial of aviotiics systems architec-
12
A.
tures (Ref 49), the Honeywell Systems and Research Center
explores the implications of fiber optic interconnections
and intelligent (i.e., microprocessor-based) communication
interface units on the architecture of distributed process-
ing systems. As a conclusion, a call is made for continuous
and considerable research and development efforts in the
area of decentralized executive control before all the ad-
vantages can be realized.
A concept presented by the Forecasting Integrated Tech-
nologies Corporation (Ref 45) concludes this review of
newly proposed solutions: an advanced memory module is pro-
posed as a medium to interface between the aircraft crew and
its operational systems in such a manner as to synergis-
'I.-.tically integrate crew and aircraft into a more effective
weapon system. The paper does not argue with the fact that
the distribution of processing resources leads to signifi-
cant avionics performance increases, but advocates that the
proper application of mass memory can add more to promote
avionics performance.
Of course, there is not a single discussion in the
literature which does not address fault tolerance aspects in
the field, but good insight into the problem can be obtained
from Ref 19, 43, and 77.
* Very useful general knowledge literature is given in
Ref 7, 13, 24, 30, 34, 35, 40, 51, 70, 71, 83, and 92.
13
III. Characteristics of Avionics' Architectures
The fact that avionics data multiplexing has accumu-
lated so much attention during recent years makes any analy-
sis and documentation of current knowledge in the field a
cumbersome task, especially if the goal is to synthesize an
architecture based upon conclusions drawn on the "advan-
tages" and/or "disadvantages" of existing technologies.
However, if that is to be done, then three sets of important
issues should be defined:
-First, one has to characterize digital avionics net-
work architectures with enough generality to allow a de-
scription and documentation of different (and sometimes mu-
tually exclusive) approaches in a similar manner;
- The second set is an evaluation tool to compare the
attributes of presented techniques with respect to tho im-
pact of implemented philosophy on overall system perfor-
mances; and finally,
- The synthesis of the "best" architecture should be
performed according to the third set of selected criteria,
recognizing all possible conflicts and necessary tradeoffs,
then making decisions on a problem-by-problem basis.
It is obvious that the issues arranged in such or
similar manner are never mutually independent and, probably,
this is the reason that the analyses performed so far are
14
* - generally made based on one common or at most two (but
v interlaced) sets of criteria (Ref 19, 49, 80, etc.). It
seems, in fact, that an attempt to develop uniquely defined
* :JJand universal criterion to judge all possible solutions
* would have no useful application in the real life -- the
"loptimum" is too much dependent upon many contradictory
factors, icungthe role and the size of the aircraft.
This has been born in mind throughout this study, leading to
the final conclusion that the "optimal" architectural ap-
proach would be the one that captures as much as possible of
the existing techniques, and that is reliable, simple, and
general enough to be standardized for use in the entire
aerospace industry, but -- ABOVE ALL -- that leaves system
designers free and technologically independent to implement
the philosophy according to the particular information sys-
tern requirement, allowing at the same time a possibility for
subsequent growth of the system with relative ease.
However, the author felt that for the purpose of this
study an arrangement of the key issues should be made some-
how, to help the organization of the study. Thus, the multi-
* ~, plex system "descriptive" and "analysis" issues as the sets
of criteria have been chosen, as discussed in next two sec-
tions. The "synthesis set" is implicitly defined in Chapter
V along with the development of the proposed architecture.
15
................ ........ 4.*-.. *~4 - ..
S -. S * ~W-
Multiplex System "Descriptive Issues"
Classification of systems by this set of issues should
allow the system characteristics to be analyzed within some
combination of generic subcategories. These subcategories
are represented mainly by the system's general characteris-
tics (intended mission, topology, architecture, control
Q. method, and message protocol) and by the physical/electrical
characteristics of the multiplex bus (transmission medium
and coupling method, maximum bus distance, signal modula-
tion, and transfer synchronization method). These issues
seem to be sufficient (Ref 19, 49, and 80) for the purpose
of system description and categorization, as follows:
System Architecture. The term "architecture" refers to
the way subsystems are connected together, including the
method to control data transfer.
Bus Control. The bus control method or bus allocation
* scheme (as defined in Ref 49:7) is the mechanism which
ensures that exactly one terminal at a time has the right of
the multiplex bus usage. This issue is discussed in the
* Multiplex System "Analysis Issues" section.
- System Topology. The topology of a data bus system is
the map of physical connections of each subsystem to the
data bus. There exist two general types of data bus topolo-
gies (Ref 48:1-25): single level and multiple level (hierar-
chical). The single level bus topology is the simplest
'S 16
scheme and is the most commonly used interconnect architec-
ture today. In this approach each of terminals is connected
to a single level bus system (Figure la) regardless of the
data bus redundancy -- the use of multiple buses for redun-
* dancy in a single level system does not change the type of
topology.
SENSOR ASENSOR S SESO1
- 1 2 INTERFACINGDEVICE&
a) Single Level Bus Topology b) Multiple Level Bus Topology
Figure 1. Bus Topologies (Ref 48:1-26)
Multiple level or hierarchical bus topology is an exten-
sion of the single level concept. If single level buses are
*interconnected in a certain manner, data on one bus system
will pass to another bus system (Figure lb). Such intercon-
nection scheme can be used to obtain functional partition-
ing in a system. This is, also, an important issue and will
be discussed in more detail later on.
.1~17
The reader should be aware, that with the increase in
the total number of processors added to the automated envi-
ronment of modern aircraft, the choices of processor inter-
connection topology and methodology becomes more and more
critical. The choices adopted will profoundly influence the
information throughput, the fault tolerance, the reliability
and maintainability, and the all pervading integrity of the
entire avionics system (Ref 44:662).
Message Protocol. The message protocol is referred to
as the message exchange sequences provided, as well as to
the message formats themselves (Ref 41:12).
Transmission Medium. The transmission medium selected
determines to a large degree the error rate and power require-
ments of the multiplex system. Some types of media that
have been considered (Ref 40:230-233) are fluid and acousti-
cal waves, electro-magnetic radiation, optical links (fiber-
optics and free space), and several types of wire cables
(coaxial, tria.x±al, twin axial, twisted-shielded pair, pla-
nar parallel flexible, and twin-lead).
Fluid and acoustical links are possible and may be con-
.4! sidered for special applications. They are impractical for
use as a primary transmission medium, however, due to long
propagation delays and sensitivity to vibration.
Electromagnetic radiation guided by means of dielec-
tric waveguides has received considerable attention due to
18
4.A
Z-7 7 '.- 7- 7% 0..- -
the high bandwidth and attendant high data rates possible
with this method. Waveguides, however, are still bulky,
expensive and highly susceptible to physical damage.
Optical data transmission links continue to show great
promise by virtue of the extremely wide bandwidth obtainable.9..1 Bandwidths on the order of 100 GHz can be realized using
fiber optics. Coupling, termination, and repair of fiber
optics remains a problem at present, however.
For these reasons, wire cable has received primary con-
sideration as a multiplex transmission medium. Of all types
considered, coax, triax, twinax and twisted-shielded pair
cables have shown the greatest potential. While coax and
triax are more suited to high frequency applications (up to
100 MHz), electromagnetic interference (EMI) susceptability
and radiated noise are higher due to the fact that the
a' shield carries signal current. Twinax and twisted-shielded
pair (TSP), on the other hand, are usable at frequencies of
10 MHz and are balanced configurations in which the signal
"carrying conductors are shielded. It is apparent that twinax
or TSP cable is more desirable at the relatively low data
rates associated with most aircraft applications. Although
twinax is clearly superior to TSP in its bandwidth and at-
tenuation characteristics, TSP is more attractive in many
applications by virtue of its extremely low cost. On rela-
tively short bus lengths, such as would be encountered on a
194'
i .
~.A .~blight-weight fighter type aircraft, the slightly higher
attenuation of TSP is not significant.
Coupling Method. Coupling consideration is another
important issue in the physical/electrical characteristics
set and the same reference as cited above (40:230-233) is
probably the shortest but still complete discussion that can
be found about the matter.
Several methods are available for coupling the users to
the multiplex transmission link:
- Direct coupling, which is a simple and inexpensive
method, but offers no isolation and presents a matching
problem;
- Capacitor coupling, that provides DC isolation, but
must be bandpass filtered at higher frequencies to prevent
ground loops;
- Transformer coupling, that maintains isolation, sim-
plifies matching, and is relatively nonlossy, making it the
* primary coupling method today. One of the disadvantages is
that the transmitted signal must have no DC component,
but this problem can be easily overcome by judicious selec-
tion of the modulation method.
The coupling problems affect the system topology (the
transmission link configuration - Ref 40:231) and could well
be one of the most important considerations in the design of
a multiplex system, especially where the transmission link
20
'" X will be subjected to a high-risk environment. Two basic con-
figurations possible are the lossless and lossy configura-i
tions. The lossless line is a classical approach to the de-
sign of a transmission link. Its basic configuration is as
shown in Figure 2. By terminating the line in its character-
istic impedance (Zo) and properly controlling the terminal
impedances, reflections are kept toaminimum. In the receive
mode, the terminals are held at a relatively high impedance
thereby minimizing their effects on the line and in the
transmit mode are brought to a low impedance.
-Ii
- Z>>Z0 (RECEI YE)2(Z 0( TRANSIIT)jI jj
Figure 2. "Lossless" Line Configuration (Ref 40:231)
The line is "lossless" only in that an attempt is made
to keep losses at a minimum and no intentional losses are
introduced. Since losses are relatively low, however, the
. .lossless line does require less transmitter power to achieve
21
aN,
. . * .-..
,-.:. a given S/N ratio in a worst case condition. From a purely
analytical viewpoint, the lossless line is superior to the
lossy line in weight, cabling required, minimization of re-
flections, hardware simplicity, and the ability to easily
increase the length of the line.
The major disadvantage incurred through the use of the
lossless line is its inability to withstand line faults. As
can be seen in Figure 2, a line-to-line short at any point
places a direct short across the transmitting terminal and
the entire system is lost. Should the line open, a severe
mismatch would occur and perhaps system would be lost, too.
The only means of protection against line faults with
the lossless line is to provide multiply-redundant lines to
whatever level is required to ensure an adequate probability
of mission success. Redundant lossless lines, however, re-
"-. quire active switching elements to select the appropriate
line, thereby adding hardware complexity.
The basic lossy line configuration is shown in Figure 3.e.
By providing loss at each terminal and looping the line,
a configuration is formed which is significantly more immune
to line faults than the lossless approach. Configured in this
manner, the line can tolerate one fault, either open or short
anywhere on the link without loss of any of the terminals. If
two lossy lines are connected in a dual-crosscoupled configur-
ation as shown in Figure 4, both lines can withstand several
22
5-
' . . . - .- . .- .- .o - .- .o - .- . .
faults without the loss of the multiplex system.
-4T
.5w
Figure 3. Looped Lossy Line Configuration (Ref 40:232)
-' - -I A -I
LINE I
4' REOT REMOTET-ERMINALj TERMINAL
____ ____ ___ ____ ___ ____ ___ LINE 2 _ _ _ _ _ _ _
Figure 4. Looped Lossy Cross-Coupled Configuration(Ref 40:232)
23
V
..'*- Additionally, the lossy line provides a form of "grace-
ful degradation" in that individual terminals may be lost
one at a time.
For the dual redundant lossless configuration it can be
shown (Ref 40:232) that the probability of loss of one or
more terminals due to line faults is
[i1- (IpN-i]2(l (1-P) N-
while for the looped lossy cross-coupled configuration, the
probability of loss is
N N 2 i 2 N-ir(i)(P ) (1-)i=2
o_ where "P" is the probability of one or more faults occurring
* on a line segment between two adjacent terminals and "N" is
the number of terminals in the configuration. By assuming
seven terminals, a graph can be plotted comparing these two
configurations for different fault probabilities as shown in
A Figure 5.
It can be seen that the looped lossy cross-coupled con-
figuration is considerably less susceptible to line faults.
This is an important consideration in a high-risk environ-
ment where the probability of battle damage is great.
Several disadvantages are associated with the lossy line
approach. Among these are increased weight, additional cable
required, and increased complexityof the line. Additionally,
24
* 4
"*. *.* ~
. . .:.,. , .o -. :,., .,L .. ,: J. .. . -. w . .. J -= ,.- ;.' . -..'. , . . . -- b. . .. -.
o, 10
L
"~ i , c, 0z
w "~
-CR08S - COUPLED
0I
W 1001.- 0
M 1 -2 - - -5 -6 -7-8 -9 -101 10 10 10 10 10 10 10 10 10 10
0. PROBABILITY OF ONE OR MORE FAULTS ON A LINE SEGMENT (P)
Figure 5. Comparison of Lossless and LossyConfigurations (Ref 40:232)
the receivers must be designed to operate over a wider dy-
namic range and more signal power may be required than with
the lossless line. A disadvantage associated with looping
the line is the fact that propagation delays may result in
the signal arriving at different times at a given terminal
due to the different lengths of the two paths around the
loop. This limits the maximum length of the line depending
on the data rate involved.
Signal Modulation. As already stated, the coupling
method and signal modulation are not mutually independent,
especially if baseband techniques are to be used to exploit
the advantage of the technique that does not involve a
carrier. In the Figures 6 through 11, the minimum transmis-* -:.?25
25
'-7--".-, " .-
- , . .a . .* A - . . -- ..
sion bandwidth (B) is expressed in terms of the signaling
rate (f ) in bps, where the signaling rate is the highest
frequency that must be transmitted to ensure that the mes-
sage can theoretically be received. The S(w) represents
power spectrum.
Of some 23 different baseband modulation techniques
possible (Ref 40:233), the seven are the main candidates for
airborne use as discussed in the literature (Ref 3:2/21-
2/25):
Unipolar NRZ modulation (a plus level represents a "1";
zero level represent "0") is the inherent form in which data
are customarily expressed in logic circuits (Figure 6). Be-
cause it is incompatible with transformer coupling, it is not
as suitable as some other methods for use on the primary data
bus. In addition, it requires twice as much signal power for
a given peak-to-peak signal excursion as polar NRZ, which
otherwise has identical properties.
Polar NRZ (same as unipolar NRZ level, except waveform
is symmetrical about zero) is easily generated from unipolar
NRZ by subtracting out its DC level (Figure 6). For this
reason, it is less suited for internal logic operations than
unipolar NRZ, but is better suited for applications on short
direct-coupled local buses because it requires only half as
much signal power for the same peak-to-peak signal excursion.
'a 26
........
J. N
%-." .'
3' 3 3, •. .
+VI0 L±L4h Now. Unipola NRZ ha a line of 0 Hz;
polar NRZ does notI I II I II 2
+V I ..... I I" "I I I B fs/2'-44:-:::..0 B- -s...
I f-V 0 /1" 2r 3/r 4/r
FREQUENCY (Hz)
Figure 6. Unipolar and Polar NRZ Modulations-3 .(modified from Ref 3:2/22)
Polar RZ modulation is depicted in Figure 7: a "1" is
represented by +V pulse and "0" is represented by -V pulse;
- * the pulses are 1/2 bit wide.
- II
S (W) 8 (sin w T/4)2
+V
, I I S(f)
0 /T 2/1T 3/T 4/T"" ""FREQUENCY (Hz)
* __Figure 7. Polar RZ Modulation (modified from Ref 3:2/22)
274.3.
• '
. . L ~ a- - - _. . .' . ' "* - ' • - - " % .... . - , . -. .- - '
.J11
4.
-{ .'.. Bipolar NRZ (Figure 8), because of its greater complex-
ity is less appropriate than polar NRZ on short local direct
4i coupled buses and, also, does not contain bit synchroniza-
tion information in the basic signal waveform ("l's" are
alternately +V or -V pulses; 1-bit-wide zero-level repre-
sents "0").
r T.. 1111 1oo1o[I V 0 0- 0
I S I I I I I I
-L -__. 1- _ fsfiIAide,
:[ 0 1fT 2/1" WT 41T-fFREQUENCY (Hz)
Figure 8. Bipolar NRZ Modulation(modified from Ref 3:2/22.)
In the delay modulation method (Figure 9), which is
..N also known as the Miller code, a "1" causes transition in
middle of bit interval; 110" has no transition, except at the.
end of bit interval between adjacent "O's".
Duobinary modulation is depicted in Figure 10: "1" is
zero level; "0" is either +V or -V, the polarity being
-4 reversed each time a "0" is preceded by an odd number of
28
consecutive "1's".
s ,,D 2 r23-2 cos(TW2 T(17 +8 cos4 wTiL
-- .22 swT-12s 3 wT+5cos2c2wT 12cos - wFo - 2I i '1 11 101 +2cos3wT-Scos-Z cT 2ca4CTI 2.-:-. +V i __ L. .~
Or r T1- -r 1 s Bf)'--Nm - _
0 l/T 2/T 3/T 4/TFREQUENCY (Hz)
Figure 9. Delay Modulation (modified from Ref 3:2/23)
" V - - - - - - - - - -
If
- i I I I I0:!1 I ,- ii 1(f sB f"5/-. _I I I I I _ I
0 IT 2T 3/T 4/TFREQUENCY (Hz)
Figure 10. Duobinary Modulation(modified from Ref 3:2/23)
29U.°
'U;
m- ~ ~ :- v:- .ve.;. ~-*.*.
.
_'% .
" *'-*- The BiPhase-Level modulation technique (Figure 11) is
also known as the Manchester-II code and can be found in
• ,. most airborne multiplex applications. In this technique, "1"
causes transition from +V to -V level in the middle of a
bit, and "0" transition is from -V to +V level, also in the
middle of the bit.
1 11 1 11 01 01 l4-- iI LE(-sin4 iT/4)
4 ~~S (Wa) 3 snwT+vInnInI n0 _fL L.I a f_
o i/T 21 3/1 4TFREQUENCY (Hz)
Figure 11. BiPhase-Level Modulation(modified from Ref 3:2/23)
A comparison made between these modulation techniques
can be used to summarize their advantages/disadvantages:
Neither unipolar NRZ nor polar NRZ is compatible with
transformer coupling. They, also, do not contain adequate
bit sync information, and are useful on direct-coupled short
local buses only in those applications where bit sync is con-
veyed by some other means (such as via a separate channel).
30
In this case, the polar NRZ is superior to polar RZ because
the polar NRZ is less complex and requires only half the
transmission bandwidth. The polar NRZ is also superior to
duobinary modulation method, because it is far simpler and
because bandwidth conservation is not a serious problem when
the cable length is short.
Polar RZ is superior to duobinary and unipolar or polar
NRZ in direct-coupled local bus applications, which require
that bit synchronization information be contained in the
basic signaling waveform.
Bipolar NRZ, as stated, because of its greater complex-
ity is less appropriate than polar NRZ on short local direct
coupled buses and, also, does not contain but synchroniza-
tion information in the basic signal waveform.
Bipolar NRZ may be more appropriate than BiPhase-Level
in cases where a separate clock channel is available and
bandwidth is a significant factor.
The delay modulation method is less appropriate than
polar or bipolar NRZ on short direct-coupled local buses
that do not require that clock information be extracted from
the signaling waveform. If such requirement exists, then the
polar RZ is more suitable.
Duobinary modulation is not appropriate for any appli-
cations that require transformer coupling, and can find the
usage when bandwidth restriction prevents use of polar NRZ
31
4.4
with direct coupling.
Level transitions in the middle of each bit makes the
Manchester-II coding inherently self-clocking, i.e., bit sinc
can be extracted directly from the data stream. In addition,
the Manchester-II code has no DC component, making this
technique suited to transformer coupling. Also, since it is
the level transition and not the level itself that is of
interest, the signal level is not critical. This simplifies
receiver design by eliminating the need for AGC, especially
when used with a lossy line configuration, so that all of
the described features made this technique the primary modu-
lation method today for aircraft multiplexing (Ref 40:223).
Synchronization. The transfer synchronization method is
the fourth issue in the physical/electrical set as defined
earlier. The bit and word synchronization must be considered
separately in the multiplexed data bus system. A separate
clock line can be used for bit synchronization; however,
considerations of reliability and clock skew (delay differ-
ences between clock and data lines) make synchronization
derived from the bit stream particularly attractive. "Self-
clocking" modulation methods such as Manchester-II or polar
RZ provide clocking information as an integral part of the
bit stream, since a transition occurs during each bit time.
For this reason, the use of Manchester-If modulation means
that no central clocking information need be provided inde-
32
4. ,V '''<<. 9 . v ,.' .< -'v -'.?'- -. '''... .'%.. '''"'"-." .
$ pendent of the bit stream for data synchronization purposes
(Ref 52:24).
The synchronization signal which occurs at the begin-
ning of each word has three essential functions: (1) it sig-
nals the beginning of a new word (or message); (2) it iden-
tifies the word type; and (3) it resolves the bit synchroni-
zation clock phase ambiguity that is inherent in the Manches-
ter data code (Ref 75:24). This word synchronization can be
provided by unique bit patterns or unique waveforms in the
bit stream. Unique bit patterns are restricted to applica-
tions where it can be assured that a particular pattern will
never occur in a normal data stream. Unique waveforms may
differ from the standard data stream in amplitude or time of
transitions. In systems using Manchester coding, the ampli-
P tude of the signal is not important; hence, word synchroni-
zation can be accomplished by using a nonvalid waveform.
This nonvalid waveform is characterized by having only one
level transition in three bit times, as opposed to the
normal Manchester level transition during every bit. This
method is discussed in more detail in Chapter IV (sections
* describing the MIL-STD-1553 based multiplex buses).
A central clock allows for synchronous operation be-
tween all the bus terminal units, but problems of reliability
.1 (a failure of the central clock disables the entire system)
and clock skew (requiring compensation logic in the terminal
4., 33
units) make this approach less desirable than the use of
local clocks. Although local clocks increase the complexity
of the terminal units, problems of clock skew and dependence
on a single clock are eliminated. This increases the flexi-
bility of the data bus system, which is particularly impor-
* tant in vehicular systems where growth and modification over
the system's life is anticipated (Ref 52:27).
Multiplex System "Analysis Issues".
This set of issues is intended to be a criterion to de-
fine the "optimal" architecture and evaluate the attributes
1/ of different techniques presented in the next chapter.
The set includes bus allocation scenarios and patterns;
bus efficiency and utilization; bus bandwidth (data through-
'S put) considerations; message latency and overhead; impacts
of word length and message addressing limitations; capabil-
ity of a system to transfer periodic, aperiodic, and/or
variable length messages, as well as possible block-messages
transfer; system integrity and error handling (recovery
schemes); functional partitioning and level of isolation;
and finally, software interface issues (software and hard-
N ware complexity trade-offs and costs). These characteristics
will be used -- not necessary in that order -- to evaluate
different techniques as described in the next chapter, and
to define the characteristics of the "optimal" architecture.
5. 4.-.34
Bus Allocation Scheme. The serial data bus is a re-
source shared by all subsystems connected to it, and, as al-
ready has been said, the method used to prevent simultaneous
requests for use of the bus (or to resolve any contention if
allowed to occur) is the integral characteristic of the sys-
tern architecture. This characteristic, namely, will deter-
-~ mine to agreat extent whether the system belongs to the cen-
tralized, federated, or distributed multiplex system archi-
tectures, that will be discussed in the next few paragraphs.
* However, it is necessary to distinguish bus control and bus
allocation mechanisms (Ref 20:46) -- bus control provides
control mechanism to establish the orderly transfer of data
from sources to sinks, while bus allocation associates the
"how" (method) and "when" (timing) of message transfers.
Some inconsistency is noticeable throughout the litera-
ture in how different authors have defined the conditions
characterizing the three types of architectures -- there are
several systems that are classified by different authors as
-4- belonging to the centralized, federated, or distributed
multiplex systems, but the authors were referring to the same
architecture! For the purpose of this report the definitions
for centralized and distributed multiplexing systems from
the reference 49:111-112, and for the federated systems from
the reference 20:48 are accepted, as follows:
35
The conventional, centralized data multiplexing systems
are characterized as having:
- Centralized, master/slave control;
- A small number of processors (often more than one
only for redundancy);
- A larger number of data sources and sinks (typically
sensors and actuators), many being quite simple;
- An interconnection (e.g., bus) structure for connect-
ing the data sources and sinks as I/O devices to the comput-
4 er(s), sometimes through multiplexers;
- Low complexity communication interface unit hardware;
- A high ratio of application I/O data to multiplex
system control information on the interconnection path(s).
Federated control is normally associated with synchro-
nous bus control which is allocated by the master computer.
The master computer contains a master executive which is
responsible for the intertask synchronization and communica-
tion external to an individual processing element (PE). The
individual PE need not know the location of any external
task since all of the control is through the master and all
communication requests must be directed to it. Time sequenc-
ing is also controlled from this central controller. In
each PE there must be, however, a local executive which is
responsible for the coordination of tasks local to a PE; the
sophistication of the local executive should reflect the
.4 36
4,
--. size of the processing elements and the complexity of task
interactions.
The list of important characteristics of a distributed
processing architecture is:
- Decentralized, cooperative control;
- More processors (on the order of 10 to 100 in future
aircraft);
-An interconnection structure (e.g., bus) for connect-
ing the processors;
- Higher complexity communication interface unit hard-
ware;
- Data sources and sinks that are interfaced to proces-
sors (perhaps via data multiplexing subsystems), not direct-
ly to the subsystem interconnection paths(s);
- A lower ratio of application 1/0 data to system
control information on the system interconnection path(s).
Distributed systems are coming into increasing use
4 throughout the digital processing field because of the flex-
ibility, performance and reliability advantages which they
offer (Ref 46:1) -- examples of benefits in flexibility are
(1) the ability to route computing tasks (or bus control,
when necessary) to the most suitable processor (as contras-
ted with local processor that may not be very efficient for
a given task), (2) the ability to add processors increment-
ally as the computing load increases, and (3) the ability to
- 37
% .~-..'-* -1- ~ .- ~ .- ;-~ * . - * p * * * 7,
.~ introduce technical advances gradually, one processor at a
time, while retaining existing computers on-line, thus avoid-
ing the major software and systems problems that arise when
.3 dedicated computers are replaced, new are added, or some are
removed from the system. Performance is enhanced because
distributed system bus allocation schemes generally require
less overhead than those in the other techniques, thus
increasing data throughput, as well as throughput of comput-
ing tasks by allowing any temporarily idle computer to be
utilized for sharing the load at a busy site. Similarly,
reliability is improved because the possibility of single-
point failure (in the bus controller) is removed and other
processors can be utilized to take up the load of a failed
processor until it is repaired. The reliability advantages
of distributed systems translate directly into survivabil-
ity, perhaps the most valuable attribute in any avionics
system, so that the price paid for by greater complexity of
"smart" bus-interface units is undoubtedly well justified.
In contrast, the primary disadvantage of the command-
response multiplex system philosophy is the fact that abso-
lutely nothing happens on the bus without express permission
or invitation of the bus controller and then only in compli-
ance with a strict and rigorously enforced set of rules (Ref
90:264) -- any desired flexibility in data rates, data flow,UK;. data selections or variations as a function of mode must be
38
resident in the bus controller. In addition, a significant
amount of bus capability is consumed by the bus controller
to command the remote terminals. On the other hand, the ob-
vious advantage of this philosophy is due to the fact that
the most complex subsystem is often assigned to be the bus
controller because it has the most capability as well as the
most need to have information from other subsystems, so that
the approach ensures that central processing is being done
with timely data by interweaving the sequence of bus trans-
action with the computer processing requirements (Ref 52:34).
In decentralized control schemes (often called "conten-
tion" protocol), all participants on the bus are usually giv-
en control of a small segment of the traffic, normally that
part which is relevant to the controlling terminal. Most
contention protocols are source oriented, that is they are
based upon transmitting data that is available, not upon re-
ceiving data from another participant; however, this is not
always true since in some systems a terminal can contend for
bus time to request data that is required. There are three
basic types of contention systems: Time Slot, Polled and
Pure Contention (Ref 90:264-265).
In a Time Slot Contention system each terminal is
assigned a fixed time interval during which it is allowed to
transfer data or control bus traffic. This protocol is very
rigid and it requires absolute synchronization of terminals.
39
In a Polled Contention system there is a system of
"Offer" and "Acceptance" based upon some pre-established
criteria. This system is semi-rigid, can be implemented in
several ways, and can utilize a variety of criteria to
accept or reject the bus when offered. Polling of the bus
participants can be accomplished in a pseudo-random fashion
by a master polling controller, or can be passed from termi-
nal to terminal in a fixed pattern (round robin).
In a classical Pure Contention protocol, each partici-
pant on the multiplex bus is allowed to have free access
whenever the bus is not being used by another system (more
-"I details are given in Chapter IV, section "Passive Bus Sys-
.* tem"). This is a very flexible system since it allows traf-
.0 fic to flow based upon need, not upon established rates or
times. Several variations are also possible, but all are
based on the "if-bus-not-busy" philosophy.
A different concept, which also allows contention to
occur but with a very interesting resolving scheme (the
"Transparent Contention" technique), has been developed
recently by the Flight Dynamics Laboratory, WPAFB, Dayton,
Ohio. The technique is discussed in length in the next
chapter (pp. 122-135).
An alternative to multiplex bus structured systems is
the so-called switched network communication system (also
discussed in Chapter IV, pp. 111-117).
.. *5"*40
* .. Bus Efficiency and Utilization. Bus efficiency is a
measure represented by the ratio of data bits transmitted to
the total number of bits transmitted (Ref 3:B-1), while bus
utilization is defined as the ratio of the total traffic on
the bus to the maximum allowable traffic (Ref 3:B-13).
These two characteristics, when combined with message
latency considerations, give a good measure of capability of
the architecture.
Data Throughput and Bus Bandwidth. Data throughput is a
measure of the rate at which a system can exchange data
* among its terminals (Ref 49:204). As defined, this rate is
primarily a function of the transmission medium utilized and
its inherent transport delay (the bus bandwidth), the number
of simultaneously active data paths, the amount of overhead
imposed by the protocol, arnd, to some extent, the character-
istics of the bus interface unit as seen by the avionic,
subsystem software.
The dedicated, point to point transmission of digital
information at rates above 10 MHz is not anticipated in the
future aircraft systems. The limitation is based on the
constraints of signal skewing due to transmission delays
encountered in aircraft installations. The use of an optical
transmission medium for the transfer of digital information
at or below I" MHz is in general not justified since the
same function can be performed by a cheaper, lighter weight
41
electrical interface (Ref 22:47-48), excluding audio, video
and radar signals.
On the other hand, the rapid advances in optical tech-
nology and LSI circuits make the use of higher bandwidth me-
dia an attractive solution. Along the psosibility to include
audio, video, and radar signals in standard multiplexing
techniques, the use of the higher bandwidth media allows in-
creased intelligence in the multiplex subsystem communica-
tion interface units. Namely, this combination of bandwidth
and communications interface unit intelligence is highly
synergistic (Ref 49:1) -- the bandwidth can be used to
achieve significant architectural advantages of distributed
processing networks, but the intelligence is required to use
that bandwidth; intelligence can be used to achieve signifi-
cant architectural advantages, but this requires high com-
munication bandwidth.
Another class of advantages which can be expected from
higher bus bandwidth is lower avionics subsystem software
cost. One example (Ref 49:88) is that the higher bandwidth
permits increased communication rates among software modules
in different computers. Contemporary software engineering
principles recognize that proper partitioning of software
can significantly decrease the cost of software checkout,
verification and maintenance. Because of bus bandwidth limi-
tations, current practice is to partition software along
42
-. lines of minimum intermodule communication. However, the
* ,software partitioning techniques which have proven to be
most beneficial tend to result in rather high intermodule
communication. Thus, they have been largely restricted to
multiprocessor systems where the shared main memory provides
high interprocessor communication bandwidth. With fiber
optics, distributed systems may begin to take advantage of
the software cost savings resulting from these partitioning
. methods.w,
Anyhow, the bus bandwidth is at a premium in systems
with many processors, as stated in Ref 55:21 describing the
transparent contention system.
Message Latency and Overhead. The definition for mes-
sage wait time, or latency, is accepted from Ref 81:21 as
the amount of elapsed time from when a message becomes ready-.5
to transmit in one device to the completion of receipt by
the second device.
Obviously, overhead imposed by the protocol cannot be
avoided in any multiplex system. The synchronization bits,
interword and intermessage gaps, bus allocation bits, acknow-
ledgments (including status words), message headers, check-
sums, etc., are the overhead.
Word Length. Digital words that are used to transfer
information in avionic systems include a variety of formats:
from complex groupings of bits down to a single bit repre-
* 43
S-'
.''-.5'.5., : ..) -< -, . , . . .,, , ", , .-, > o-,,, ,,', .; - '" . ... . . ' " - ' " " "' " " " " " " " ' " " ' '" "2. " . <
-. senting the state of a discrete switch position. For this
report, the definitions of types of digital information that
need to be considered for avionic subsystems are accepted as
follows (Ref 22:56):
Control Word - digital words that require one or moreactions to be taken by the recipient;
Data Word - a digital word that contains informationrelevant to the operation of the recipient;
String Message - a group of digital words which containinformation relevant to the operation ofthe recipient, but for which individualwords contain no useful information;
Block Message - a group of digital words where each wordcontains useful information, but the entiregroup of words is required in order to berelevant to the operation.
The data word length, defined as the number of bits in
-~ the word, determines the accuracy with which the processor
* can readily perform any calculation. When a word length of
* 'In" bits is used, a number can be represented to an accuracy
of one part in 2 . If the processor is not to degrade the
accuracy of the system, the word length must be consistent
with the accuracy of the transducer data entering the sys-
tem. In general, avionics data is seldom as accurate as 0.1%
* (Ref 13:290), and so 10 bits would be adequate. The custom-
ary 16-bit information field in the data words is found in
S most of today's avionics multiplex systems, allowing further
accuracy (including for round-off errors in calculation).
44
. - - o -- -
. - . , * * , , .. *° - . - . ~ 2 .- *. . , ;. .. , *'• -
The actual data word length is often 20 bits long and
includes timing synchronization and word parity, in addition
to the useful 16-bit information field. Once determined in
such or similar way, the data word length determines the
length of the other words in the system (control word,
status word, etc.) due to the necessity of unified words in
I given system.
Message Addressing. A message can be marked by the ad-
dress of the source and/or destination(s), or may contain the
"message name" recognized at the recipient site. The number
of available bits for message addressing, as well as the
"q' addressing method utilized, impacts the system's flexibility
and capability to accept subsequent growth. The source and
message addressings are oriented toward broadcast messages,
but destination addressing is oriented toward single-desti-
nation messages (Ref 49:125-129). These facts affect the
system extensibility: to add or remove a message source
with destination addressing requires no changes in other'p". subsystems; but with source addressing, a source change
means changes in all other subsystems that receive messages
* -C. from that source.
On the other hand, a system in which the source of a
broadcast need not know the number of destinations is more
extensible than if it must know, and it is even more exten-
sible if the number may be one (i.e., if broadcast is implic-
45
..- '
it). When explicit broadcast to an unknown number of desti-
nations is employed in conjunction with messages addressed to
destinations by "logical names" (i.e., software-defined,
versus specific hardware unit "names"), then hierarchical
addressing is possible. Hierarchical message-addressing se-
lects a class of destinations based on the structure as well
as the content of the address. For example, suppose that
destination addresses are four hexadecimal digits each and
that a zero digit is interpreted as "don't care". Then the
names can be assigned such that all four-digit names without
zeros are those of (or within) individual subsystems, while
those with one or more zeros are names of groups, as shown
in Figure 12.
0000
AOOO BOO0
.. /AAOO ABOO ACOO BAOO BBOO BCOO BDOO
AB70 ABFO
A873 AB7E9..
Figure 12. Example of Hierarchical Message%%.. Addressing (Ref 49:126)
46
. =% >*.K. . ,,'** v %-%,..v %-;. .% ,.,% V% ": . -..- , .% • , . .. . .•.. ... . .. . . .
*z7 7 -T 7. 7.*.. 77*
- Thus, messages addressed to 0000 are received by all
subsystems, and those addressed to AB7E are received by it
alone.
a The obvious advantage of the hierarchical message ad-
dressing is that it matches the hierarchical structure of
1 the avionic subsystems (Ref 49:125-129).
Periodic versus Aperiodic Messages. In spite of the
fact that changes in the avionics data are highly aperiodic
in nature, most multiplexed avionics systems today operate
on fixed schedules of data transfers. This is, probably, the
consequence of the widespread use of command/response proto-
cols that inherently require scheduling of sampled-data mes-
sages that occur at regular time intervals. Handling of un-
-: scheduled (aperiodic or "trigger") messages in such systems
sometimes requires an extensive consumption of the bus
bandwidth due to polling of the terminals.
The requirements for the scheduling come from the exam-
ination of the largest and smallest minimum iterations and
allowable latencies, as described in Ref 3:3/14-3/16. The
slowest iteration rate, which is the least common multiple
.4 of the faster iteration rates, is normally defined as the
major cycle (Figure 13).
Over the course of a major cycle, all periodic trans-
missions occur at least once and all periodic computations
occur at least once. Some exceptions do exist if the itera-
a. 47
a . t - .5 J . - - .V .- -W W - . A . A -. . . ......
tion frequency is very low (such as Kalman filtering once
per 6 seconds, for example).
MAJOR CYCLE
• MINOR MINOR MINOR MINORCYCLE CYCLE CYCLE CYCLE
NO.1 NO.2 NO.X NO.N
9.
Figure 13. Major and Minor Cycles (Ref 3:3/16)
The minor cycle is normally the frequency of the most
rapidly transmitted periodic data. Typical major cycles are
about one second in length, while minor cycle lengths can be
either binary (2/ sec) or decimal (10/ sec), with common
values being 1/128, 1/64, 1/50, etc. (Ref 3:3/14-3/16).
Aperiodic message (time-critical or "trigger" message)
is normally assumed to be a very important component of the
bus traffic, although such messages are rare and comprise a
small percentage of the total message traffic. A weapon
delivery message or threat warning message are examples of
such messages.
In a command/response system the bus controller polls
the terminals at periodic intervals to determine if a termi-
nal has such a message ready to transmit. If there is one, a
48
I .,
communication link will be set up in the manner normally
used for periodic message transfers. The procedure is re-
peated as long as the terminals have these interrupt messages
to be processed. Then, the bus controller returns to the
regularly scheduled transfers. Obviously, the procedure
introduces significant overhead in the message traffic. For
example, to achieve the average delays for aperiodic mes-
sages of about 1 ms, the bus controller would have to scan
all terminals at a rate of 1000 times per second just to
look for interrupts alone. Since other periodic message
traffic has to be accommodated as well, this rate could be
burdensome (Ref 80:2-9).
The attempts made in recent yearz to design multiplexed
V avionic systems with transmission of only data that has
changed since the last update (the "information-only trans-
fer networks", Ref 61:690), resulted in significant improve-
ments of data latency, providing a meaningful contribution.4
to bus efficiency as well (Ref 90:430). The distributed
control systems involving such a transmission scheme explore
benefits of aperiodic nature of avionics information. Also,
in such systems all messages are basically aperiodic and
handling of any trigger message is the normal procedure
rather than an exception.
Block Messages Transfer. Multiplexed avionic systems
with a block (or string) message transfer capability gener-
49
%k 3 .* 5*4. * h .*.,. . . b'. ,,t ,,.% _ ' ' ' , , . ., - . . . ' , ,*.- . - . ,
- - - - - - - - - - - -n.. . - .-- t
ally demonstrate a better bus efficiency measure than sys-
tems without this feature. This is the consequence of the
lower overhead required to transmit a message block versus
word-by-word transmissions. However, the block message trans-
fer capability requires more complex hardware and/or soft-
ware support, especially when the messages are of variable
lengths. In addition, the longer the message permitted, the
* higher the probability in that a few terminals can tie up
all buses available in the system and prevent other messages
from being transmitted, unless special restrictions are
imposed by the system's hardware or software.
There is no sound evidence that would justify more
complex hardware (thus reducing reliability) and/or software
(thus increasing cost) just to have the block message trans-
fer feature. A system that requires software reloading as a
part of its recovery procedure is an exception.
System Integrity. For real-time command and control
systems such as an avionic system, the integrity is consid-
ered to be the most important aspect of a system. The
system integrity is defined as the degree to which a system
tolerates faults, errors, and failures. Faults are referred
to as mechanical or algorithmic defects which may generate
errors; errors are items of information which when processed
by the normal algorithm of the system, will produce fail-
ures; and failures are defined as the events at which a
05
K '.-~--system violates its specifications (Ref 49:203). Integrity
involves confinement, detection and diagnosis, recovery
(preventing further damage and restoring a stable consistent
system state), repair, restart, and reporting. This attri-
bute, therefore, encompasses those referred to by the terms
fault tolerance, reliability, availability, and others.
Note, however, that error handling and recovery proce-
dures are dependent to a great extent on the system's appli-
cation and requirements.
Functional Partitioning. The primary reason for utili-
zation of multiple-level topologies in today's aircraft is to
achieve high level of isolation between flight-critical,
~ flight-phase-essential, and nonessential aircraft functions.
Such functional partitioning within a data bus system inte-
gration separates the aircraft functions, so there are mini-
mum reactions of other functions based on a possible failure
of one function. Obviously, both hardware and software func-
tional partitioning techniques are integral and very impor-
tant elements of the system integrity. However, the tech-
niques of functional partitioning are treated separately in
.9 this report to emphasize the importance of the functional
partitioning characteristics of a system.
The Federal Aviation Administration (FAA) believes (Ref
85:2) that the possibility of significant design errors in
software based digital equipment is much more likely than
51
,' '.>for a comparable analog system. The possibility of undetect-
ed software errors results in serious problems when attempt-
ing to certify installations of redundant digital equip-
ment to be used for critical airplane functions. Namely,
redundant installations do not provide protection from de-
sign errors. It is believed that functional partitioning, if
implemented from the lowest to the global bus level, could
help to some extent in solving such testing problems, and at
the same time, raise overall system integrity.
Flexibility. The flexibility feature of a multiplexed
* avionic system does not include only the provisions for
anticipated growth during the system' s operational life.
Additions, deletions, or substitutions of subsystems con-
nected to the bus(es) are to be normal daily procedures to
reconfigure an avionic system in response to battlefield
changes. The system ability to accommodate widely varying and
* changing weapons, thus achieving true multi-mission capabil-
ity, is the unavoidable requirement today. Distributed,
multi-level, hierarchical processing networks with a modular
design of avionics show a great potential where flexibility,
expandibility, and system integrity are required at the same
time. An extensive use of modular design permits distributed
approaches to the architecture synthesis. This will allow
rapid, on-board verification that the reconfigured system
will not endanger safety of the next flight. The resulting
52
* .flexibility is expected to allow an aircraft to fly more
than one kind of mission in a single day (Ref 74:585).
Software Interface. The software interface is the sin-
*gle largest factor in the impact of data multiplexing and
distributed processing on avionics subsystem cost, affecting
system performance, integrity, and other factors as well
(Ref 49:117). At the same time, even conservative projec-
tions show that an increase of an order of magnitude in the
amount and complexity of advanced avionics software can be
anticipated in the years to come.
To illustrate what such a trend really means, an exam-
ple of today's software complexity and costs will be pre-
sented (Ref 74:576):
0" Assume a 12000-line avionic program was to be acquired
with either 0.5, 1.0, or 2.0 percent of the lines of code in
- I error ("instruction error confidence levels").
The acquisition cost was definedas$2,400,000 (25 lines
of code per workmonth), $1,200,000 (50 lines of code per
workmonth), and $800,000 (75 lines of code per workmonth)
for the respective error confidence levels.
The program scenario defined it as having 10 years of
useful life and being installed in 180 processors distrib-
uted in the same number of aircraft among 10 tactical
squadrons. In a squadron, 10 aircraft make two flights of
two and one-half hours duration per day.
53
2 .
The total operating time of the fielded system may then
be computed by:
1.825 x 106 hours = 100 aircraft x 5 hours/aircraft-day x3650 days
Using the definition for failure rate, and its inverse,
mean time between failure (MTBF), the following equation was
applicable:
Errors =(error confidence level) x (program size)100,000 hours (fielded system operating life)
The 0.5, 1.0, and 2.0 percent error confidence levels
gave 3.29, 6.58, and 13.20 program errors per 100,000 hours,
respectively. Using Monte Carlo techniques with a normal
distribution for error difficulty (two, four and six work-
months to fix one, two, and three sigma errors), and an
exponential distribution for time of error occurrence, the
following table was developed:
TABLE I
Typical Average Maintenance Workloading
Error Rate MTBF People Cost
0.5 30395 1.5 $ 900,000
1.0 15923 2.4 $1,140,000
2.0 7575 4.8 $2,880,000
(Ref 74:576)
" -" 54
The part of the life cycle spent on maintenance for the
three error rates are seen to be 27, 55, and 78 percent, re-
spectively. A plot of procurement and maintenance curves, as
well as a plot of their composite, is included in Figure 14.
This composite conforms to the classic life cycle curve with
its saddle point of cost effectiveness. Even when the saddle
point of the life cycle curve is used, three people are
required for the software maintenance.
-4n
2.000
CMPOSNTEAC
,00 7.00 4,002.0 800 3.0
MPROCUROMEN
-5-5
7Wt % '
The operational avionics software problems alone are
not within the scope of this study and they are treated only
as an integral part of network architectures. In that sense,
any software structure is considered in the light of its
conformity with the proposed hardware structure. For example
(Ref 80:2-5), in the case of a command/response architec-
ture, the design of all aircraft subsystems must be complete
before the data bus control program can be completely writ-
ten and validated. Otherwise, whenever a subsystem change is
made during the design phase of the aircraft's avionics,
changes must be made in the developing central bus control
program, and the entire software program must be validated.
* Any reconfiguration and subsystem alternations in an opera-
tional aircraft will have the same consequence. As shown in
Table I (page 54), software maintenance is very costly, and
obviously, any system software revalidation must be even
more expensive.
In contrast, the software development for subsystems in
a contention multiplex system can be done in parallel and
the validation can include only testing of the data transac-
tion between a source and a receiver, because there is not
exclusive control scheme resident in a single site to be
validated. The system reconfiguration has no so significant
-' impact that requires entire system revalidation. This is
true especially if data broadcasting is the only means of
U. 56
communication in the system.
The other side of the coin is the fact that centralized
systems, once operative, provide processing with timely
data, as already has been discussed (page 39). In flight
critical systems which utilize distributed multiplexing
techniques, the data transfer synchronization may become the
main problem. Software modularization, event-driven schemes,
quantized software, data time-tagging, etc. , are some of the
ways to get around the problem and will be discussed in more
detail later.
Hardware/software complexity trade-of fs are another
important dilemma. A high level of hardware support can
decrease software complexity and costs, but introduces hard-
ware reliability problems. A more advanced architecture can
offer a highly reliable system made from unreliable pieces
(Ref 49:88), but obviously this has a limitation.
Chapter Summary
In this chapter, the current theoretical considerations
from the available literature have been used to sort out the
essential figures of merit of computer network architectures
I for digital avionic systems.
Two sets of key issues have been defined: the multiplex
system "descriptive issues" and multiplex system "analysis
issues". As has been pointed out, the sets are not mutually
57
* *i * *.~*.*~~ 2* ~%4
~ .~ independent and the characteristics from each set are dis-
cussed with respect to the advantages and disadvantages that
they offer.
The characteristics will be used in the next chapter to
describe and analyze current network architectures, as well
* as in Chapter V to define the "optimal" architecture for
digital avionic systems.
.658
*~7 .' -..7
IV. Avionics Architectures Description and Analysis
In this chapter, descriptions and simultaneous judge-
ments are given on fourteen approaches to the avionics data
multiplexing. The choice of a system to be -presented is
based on either its historical importance or the signifi-
cance of the ideas included in the approach. However, it is
always questionable whether an approach is the "significant"
one or not--it depends on the designer's needs and, natural-
ly, each of the technologies does have its own merits for
the particular application considered. For complete discus-
sion on some other techniques, the interested reader is
referred to the bibliography. (For example, the multiplex
systems for F-18 Fighter/Attack Aircraft, YAH-64 Advanced
Attack Helicopter, and B-52's Offensive Avionics, are de-
scribed in Ref 41; the Space Shuttle Orbiter and the Advanced
Medium STOL Transport Aircraft multiplex systems are de-
scribed in Ref 49; the Light Airborne Multipurpose System
(LAMPS - a Navy ship and helicopter system) is given in Ref
3; the Advanced Integrated Display System - AIDS can be
found in Ref 67, etc.).
The parallel bus systems are not considered as suitable
as serial multiplex bus for avionics systems, and will not
be discussed in this paper for the same reason, although
the parallel bus has the potential for being much faster
59
\ -'..y.than a serial bus. Namely, in a parallel bus system, the
number of interconnected wires could be very large and could
greatly inhibit the degree to which the subsystems could be
physically distributed. Also, a failure in a single wire of
the parallel bus would essentially cause the entire bus to
fail, while any redundancy requirements greatly increase the
hardware and software complexity.
The descriptions and analyses for the chosen systems
are kept as short as possible. More data can be found in the
reference(s) given at each section title.
The F-15 Multiplex System (Ref 49:13-20 and 44:41-48)
The F-15 is an Air Force air superiority fighter manu-
factured by the McDonnell Aircraft Company. An avionics data
bus system is used to exchange data between various subsys-
tems, most incorporating a digital processing capability.
The multiplex system is unique among the systems consid-
4. ered in that it predates all aircraft multiplex standards.
4% The system architecture is illustrated in Figure 15, which
also lists the attached subsystems.
This single-level topology employed in a federated ar-
chitecture contains a single central mission computer (MC),
that controls two sets of dual buses. Backup buses for crit-
ical functions are depicted by the dashed lines in Figure
X4? 15. Note that flight control is not on the bus, because it
-V ~ 60
. . was considered to be too critical to place complete confi-
dence in a multiplex system.
---------------------------------------
Ir-MJA -N -" --IRaw U NVle4ATIiN
I It, IIU
i--- ATTTM U iA UL4
4,1 39CNMtC O"I I-UA D1l ATA
L , ILT
MADI 'IIIN
I-------------------------. . . .... I
Figure 15. The F-15 Data Bus Architecture (Ref 41:44)
Transmissions are half-duplex (two way but not simulta-
neous) and are controlled by the MC. A timing clock from the
MC is supplied along with the data. The reference clock
signal is also used to identify which of two redundant buses
is to be used for data transmissions. When the MC wishes to
use one of the two redundant buses, the reference clock
signal is brought up on the clock line of the selected bus.
The two (shielded-twisted) pairs of data and clock lines
61
-{:.~.*-.constitute a bus, having a characteristic impedance of 68
ohms each. Data is transmitted on the bus at 1 mbs data rate
(Manchester-II coding is employed). The signals appearing on
the transmission line very closely resemble 1 MHz sine waves
since the harmonic content of the data is limited above 1.5
MHz. Data signals, when all either logical one's or zero'ss
look like 1 MHz sine waves. Data signals of altering logical
one's and zero's are 500 KHz sine waves mixed with 1.5 MHz
3-rd harmonic components. This produces a wave form with a
* - zero transition which very closely approximates the zero
* transition of a 1 MHz sine wave. Clock signals look exactly
like data signals of all logical one's; i.e., a 1 MHz sine
wave.
Maximum line length is considered to be 75 feet as
limited by the acceptable phase distortion of the separate
clock line in the noise environment of an aircraft installa-
tion. The receiving branches of the bus network are termi-
nated by the high input impedance of the terminals (5K ohms
resistive component at 1 MHz). No transmission line termi-
nating resistors are used at the ends of the line. Multiplex
a~aaterminals are transformer coupled to the line in a balanced-
to-ground, center-tap grounded configuration. In the trans-
mitting mode, the source impedance of the terminal is 68
ohms. Since the total transmission line length is a small
V portion of an electrical wave length at the primary frequen-
a~a 62
* - cy of 1 MHz and frequencies above 1.5 MHz are attenuated
prior to transmission, the primary reflections do not pro-
* duce significant signal waveform distortion.
Message and word synchronization is determined by exam-
ining the blank transmission times. More than eight clock
periods without data constitutes the end of a message. Words
within a message are separated by five clock periods.
Data messages (Figure 16) consist of a select word and
1 to 15 data words, each word containing 16 bits of data and
1 parity bit.
The select word originates only in the MC and performs
one of the following functions:
- Request data transmission from specified hardware;
- Identify data to be transmitted from the MC to a
specified peripheral;
- Command a peripheral to take some action other than
transmit data.
The MC communicates with each subsystem on a one-to-one
basis, as specified in the select word address field. Addres-
* ses are defined by software contained in the MC. Terminal
addresses are implemented in hardware. A control field spec-
ifies the message type or command action to be taken. A
* command indicator bit specifies the select word as a command
word, while another bit specifies receive or transmit.
* Data transmissions are not acknowledged, but command
63
messages are. The receiver returns the select word as the
data word to the MC when a command is issued.
The word count field in the select word (one for mode
commands) indicates the number of words to be transmitted.
SELECT WORD FORMAT
UNIT MESSAGEADDRESS CONTROL FIELD WORD COUN
10 1 2 31 41 S 61 71 8 9 10 111 213141 1
COMMAND TRANSMIT/RECEIVE PARITYINDICATOR. BIT BIT
DATA WORD FORMAT
0 1 23 1T41N 16T17T 1P19110111112113111
a-,DATA TO BE TRANSMITTED PARITYBIT
MESSAGE FORMAT
DPsDPDDPDDPD [] PS* LAST SELECT FIRST SECONDLAST NEXT
DATA WORD DATA DATA DATA DATAWORD WORD WORD WORD WORD
PS- INTERMEDIATE SAP TIME
PD INTERWORD SAP TIME
oz I CLOCK PERCOMP00 5 -CLOWK PGRIOOS
Figure 16. The F-15 Multiplex System Wordand Message Formats (Ref 49:17)
Bus traffic is completely controlled by the MC using
command/response techniques. The MC performs mission orient-
ed computations and either uses or generates most of the
64
--,-p €,,,e ,.., -,,.,... -.....-. ? ... ,..:"', '. . . - ,b :"""' "-/ "- - L- ". . "-" j."."
~.. data on the bus; there is very little requirement for data
exchanges directly between remote terminals over the data
buses and no transfers between remote terminals are provided
at all. If the MC wants to regain control of the bus due to
format errors (invalid data bits, parity errors, incorrect
word length or spacing) recognized on the bus while a termi-
nal is transmitting, the transmitter will be shot down by
dropping the clock signal, which will cause all the RT's to
go to the receive mode, resynchronize and prepare to receive
a new select word.
A disadvantage of the solution is inherent to the com-
mand/response technique utilized: all transmissions on the
MC bus are synchronous with allocation via table lookup in
the MC software. The consequence is that in order to change
the allocation scheme, including the number of terminals,
the MC software would have to be modified.
Also, the number of terminals itself is limited by only
4four bits available for the terminal addresses (to 2 =16
terminals).
There was no actual data available, but for the purpose
of this study the bus efficiency is calculated to be from 13
to 68 percent, depending on the number of data words in a
message. It is assumed that the select word, command acknow-
ledgment, parity bits, and intermediate and interword gap
times are all the overhead. Bus utilization is around 10
.%.. 65
percent (Ref 49:15), while message latency can be up to 2.1
milliseconds, provided that all terminals interfaced to
either set of dual buses send full sequences of 15 data
words without any command word-sent in between (again, the
actual data was not available).
The MC has no knowledge of bus usage requests from any
terminal.
Synchronization loss, dead terminals, and illegal se-
quences are all possible bus control errors and will be
detected by the receiving terminals. If a receiving terminal
detects an error, it ignores the data and resynchronizes to
receive a new message. The MC checks word length, word
count, interword gap, data dropout (Manchester code viola-
.0 tions), and parity. The MC may retry transmission on either
the same bus or the standby bus, depending on the past
performance. The switch to the standby bus is only upon
fault detection on a primary bus and is controlled complete-
ly by software.
Subsystems are redundant only in the bus interface
circuitry. The interface circuitry is simplified by the
added synchronous clock reference at the expense of added
aircraft cabling and limited bus line length.
The B-i Multiplex System (Ref 49:30-35 and 36:83)
Once the data bus concept had been proven on the F-15,
66
another major USAF system, the B-i strategic bomber built by
the Rockwell International, designed its avionic system
around the similar data bus (Figure 17). The B-i's design
philosophy was greatly impacted by the MIL-STD-1553 being
prepared at the same time.
Three separate bus systems are utilized in the B-i
avionics system, representing a multiple-level topology with
equivalent levels of control. Such topology is the primary
reason that this system is described here. The first bus,
quad redundant, is used for electrical power switching con-
trol and management (EMUX); the second is dual redundant for
both offensive and defensive mission avionic sensor integra-
tion (AMUX); and finally, the third is nonredundant data
bus for built-in-test purposes (CITS).
Interconnections between the three bus systems are
provided for data exchange purposes only. As with the F-15,
flight control remained independent.
The data buses consist of a single twisted shielded
pair, with the Manchester-II modulation at 1 mbs rate. Line
length is limited to 300 feet. Explicit word synchronization
is provided by a three microsecond, 50 percent duty cycle
square wave in the same manner required by the MIL-STD-1553
(described in the next section, beginning on page 72).
Data transactions are one-to-one in a command/response
protocol similar to the MIL-STD-1553.
67
-. 4. . . • ° . . . . o . *
AVIONICS CONTROL UNIT COMPLEX
COMMONNAY A WEAPON DELIVERY
I' - MEORY - * INERTIAL SET 0 1Ii * INERTIAL SET 9 2
*L
2-'I
I .ELOI DISPLA ONO * TFRON TTO
NIT DEESVVER * ROPEAR ON:. =-- DFNI PODER PNTETE 1
Ii . "INRCRXAF.T
, .4
== PARALLEL EOMMUNICATIONPATHS
SRADAR ALTERNATE 0 2
STORES MIANAGEMENT
DEFESIV MAAEMN SYTE NUCLEAR S.U
DCONTROL / DISPLAY•DEFENSIVE SYMIBOL GENF.ATOR
ELECTRONIC DISPLAY CONTROL FRONT STAT70NTDEFENTAVE INTEGRATED K T SYSTEM""" • ¢~RAPHICS GENF.RATOR •RA TTO
:--.'. •DEFENSIVE POWER PANELEXC EXE CANTROLNEERFR OA L I IRAIRCRAFT
SLU-S A CITTFR -,ERRAIULTIPLE YSTEM
ECMSUSYTE MISSION 6 TRAFFIC CONTROL
AC l
mE =_==PARALLEL COMMUNICATION PATHS.. DEFINITIOIS4
FigACU - AVIONITA CONTROL UNIT
',.- 8LU - STAT ION LOGIC¢ UN IT -
.,TFR - TERRAIN FOLLOIN RADAR
:.,,Figure 17. The B-1 AUX Bus Architecture (Ref 49:31)
68
L. °''
.*. ..; In the AMUX (describing here only one out of the three
buses -- Figure 17) there are two identical Singer-Kearfott
SKC-2070 computers: for the General Navigation Avionics
Control Unit (GNACU) and Weapon Delivery Avionic Control
Unit (WDACU), connected in such a way that each computer
treats the other as a subsystem interfaced to the buses
under primary control. Each computer retains a complete
executive in memory such that either may assume the full
load upon failure by the other. Each ACU has the primary
role to control one set of dual buses.
Word length is 24 bits (Figure 18). Data portions are
16 bits plus one odd parity bit. There are three blank bits
and one validity bit to indicate the detection of a message
error.
-; The command word contains the subsystem address (five
bits) and a 10-bit mode/control and word count field that
allows up to 1024-word block transfers to be effected under
software control. The subsystem responds with an echo of the
command word before transmitting data or after receiving
data. Subsystems must use a data word to provide status
information to the ACU. Addressing is fixed in subsystem
hardware.
Mass memory interfaces to the ACU's through a parallel
. bus (dashed lines in Figure 17), which requires about a 1.6
megabit transfer rate.
,-"-69
.- ~ - '7... . .
'
j ~1.0pLSEC.11315 789O222
,I1213 14IJ516171 [911 14 A 1311411511 171191l912012112 2
SYNC 8 B SUBSYSTE T/ SUBSYSTEM NUMBER OF P
ADDRESS DATA BLOCK WORDSI / OR MODE
po I -LCONTROL
-'j SEC COMMAND WORD FORMAT
:li: l a I '11.1I141I5161718 19 4o ( 'ls* 14 A1145117118l19FlTol21~: L SYN,*I1 e SSYTT/' SUBSer I P BoL
ADDRF-$i DATA BLOCKL WORD/ OR MODE- B.L J..LCONTROL .
RESPONSE WORD FORMATI 1 I sl6171Ie91 4 (11', 14 A 13114115* q 17181112012, 1 12"
. -'.DATA LEOMEN-"--CNTROL BIT-. "',V-VALIDITY BIT
, DATA WORD FORMAT BBAKBTI -! T/R-TRANHIT/RECEIVE BIT
P-PARITY BITNBB-MOST 1|NIFICANT BIT
Figure 18. The B-i Multiplexed Serial DataWord Format (Ref 62:208)
*. The protocol is rigid as much as in any command/
[: response protocol. Software changes are necessary at the
bus control units if new terminals are to be added. A logi-
cal limit of up to 31 terminals is the consequence of only
5 address bits available in the command word format.
aBus utilization is up to 45 percent (Ref 49:30) and ac-
.- tual bus loading or response time (message latency) are not
70
PN
considered critical. Bus efficiency depends on the number of
words within a message and is calculated for this study to
be from 22 to 66.5 percent.
The advantages of the solution are the possibility for
block messages transfer and the fact that in order to accom-
modate a message with a higher priority, any transmission on
a bus can be interrupted by a signal over the alternate bus.
The anticipated errors on the bus (invalid sync code,
illegal word or message lengths, parity, invalid terminal
address, and data dropout) are all checked by the terminals
and by ACU's. Any error that is detected results in no
action. When a terminal detects an error and responds with
the validity bit set in the response word, or if there is no
response by the terminal at all, the ACU will repeat the
*, same message on the alternate bus. If the retry is success-
ful, further transmissions will proceed on this bus until
the next failure. Several retries are possible, or the
control will be passed to other ACU immediately, depending
on the nature of the message. Declaration of the failure is
provided, as well.
Failure of an ACU causes the other to assume the entire
load. Both ACUs are of equal capability and both have the
same programs resident in their memories. Data table ex-
changes between ACUs are performed each major frame as a nor-
mal procedure, thus virtually continuous processing is main-
71b- - - - -7, l
*i*
tamned even during the switchover due to the ACU's failure.
The F-16 Multiplex System(Re 36:84, 44:15-20, and 49:21-25)
The F-16 is an Air Force combat fighter supplied by the
General Dynamics. The F-16 development program coincided
closely with the initial publication of MIL-STD-1553 and so
became the first vehicle to utilize and flight test a MIL-
STD-1553 compatible multiplex data bus system. Due to its
extreme simplicity, it is often used in the literature as a
4'.. descriptive example of the classical implementation of the
MIL-STD-1553 standard.
The avionics architecture of the F-16 may be described
as a federated multi-computer network (major subsystems have
data processing and memory capabilities), although only a
Delco M362F computer -- the fire control computer (FCC) is
4. used as a general purpose processor (Figure 19).
The FCC operates as primary bus controller (BC) for the
dual buses and sends a separate bus control signal to the
inertial navigation unit in case of the FCC failure. The
inertial navigation unit is then responsible for bus con-
trol, providing a backup system with limited capability.
The flight control function is still absent from the
bus. The Westinghouse fire control radar is also separate,
using an internal digital multiplex data bus.
72
% .
FIRE CONTROL FIRE CONTROL BUS CONTROL INEcRIL HED.NAVIGATION COMPUTER I NAVIGATION DISPLAY UNITPAN (FCFN) (FCC) DISCRETE UNIT CINU) c.UD)
T. I i " . I 1 i
TARGETSTIRE COTROL CENTRAL AIR RAAR E/O IDENTIFICATION
MANAGEMENT DATA COMPUTER DISPLAY SET LASERSET (SHO) F) CAD) CREW (TISI)
Figure 19. The F-16 Multiplex SystemArchitecture (Ref 41:17)
The data bus itself consists of a dual set of twisted
shielded pairs with characteristic impedance of 70 ohms. Al-
though the MIL-STD-1553 allowed up to 300 feet of cable, the
F-16 data bus utilizes a short cable assembly (17.5 feet
long). All subsystems ("remote terminals" - RT) are attached
to the bus by the use of stubs which are connected to the
main bus by transformer/resistor coupling networks. Digital
transmission is at one megabit/second using Manchester-II
code (Figure 11, page 30).
Word synchronization (Figures 20 and 21) corresponds to
73
" .r. ,-, - , >,.,,,-. -,- . * ,,--. , *-... .,. .-- ... , ,. . .... .. €.. . '.;a , G'"; M'., . . ;a", l"
- **. * ..
MIL-STD-1553. The self-clocking feature of the Manchester-II
code waveform is used so that there is no need for a sepa-
rate clock line (in contrast to the F-15 approach).
+ VOLTS rI I I " 1,17
I
- VOLTS
DATA WORD SYNC DATABIT SIT
0 Figure 20. The 3-bit Illegal Manchester-II Sync forCommand and Status Words (Ref 65:7)
~~~+VOLTS - -II I
0°,I I I
- VOLTS
DATA WORD SYNC DATABIT BIT
Figure 21. The 3-bit Illegal Manchester-II Sync.. for Data Words (Ref 65:7)
74
". " " . . " " A " "" " " " " ' " , , - ", " " " " " - " "% ' ' " ", ,' . . . ' " -
Three types of words are used to make up the messages
which are transmitted between subsystems on the data bus:
Command, Data, and Status words, as shown in Figure 22 with
the bit assignments depicted, too.
Transactions are command/response with BC-RT, RT-BC, and
RT-RT exchanges. A bus controller to remote terminal trans-
fer is accomplished by the bus controller sending a"receive"
command word followed contiguouslyby the data words. After a
specified gap time, the addressed terminal responds with its
status word, thus indicating proper receipt of the message.
Command words are only transmitted by the bus control-
ler, status words only by remote terminals, and data words
by either BC or RT. All words are preceded by a unique
-0 synchronization waveform (Figures 20 and 21), which is dis-
tinguishable from the remainder of the normally Manchester-II
encoded bits.
A terminal to controller transfer is initiated by the
controller sending a "transmit" command word to the remote
terminal. After the gap time, the addressed terminal re-
sponds with a status word contiguously followed by the speci-
fied number of data words.
The direct terminal-to-terminal transfer may be affect-
ed by a combination of the two previous commands, involving
two command words and two status words.
~-" 75a. ,
• p., . . .,. . . .. ''". . . .' ~ ""-'"" '. /
'K
SIT TIMES 1 2 13 1' 1 6 17 18 19 110111112113114 15 16117 18 19120
COMMAND WORD F
SYNC TERMINAL SWADOESS/IOOE DATA W=C OWSM PADDRESS o
DATA WORD 16
'4 SYNC DATA P
STATUS WORD E:S 1 i '
. . m SED ON Y NC BYF¢FR TERMNAL ZTATU , NORATO,, 0 " M1
SYC ADDRESS 1 0-
Fiur am) vol. F-16' WodFra% R 3C . )
... *
y r M --4 -I30 o"mnm -- ra "n n -
44oOT 0 A
SUSED ONLY BY FCC FOR INTERNAL STATUS INFORMATION. I
Figure 22. The F-16's Word Formats (Ref 3,Ch.6:6)
-4 76
i7,
Addressing is mechanized in hardware within the remote
terminals. The subsystem is capable of receiving a command
word on either bus at any time: if a command is received on
the second bus while a message is being received on the
first bus, the subsystem invalidates the first message and
switches over to the second bus. Message length is specified
by a word count in the command word. Messages may be from
one to 32 words long.
Other subsystems cannot request interrupt servicing
except by a status word during scheduled servicing. All
transactions are scheduled under software control with vari-
ations allowed depending on mode indications.
The bus control algorithm consists of about 1800 words,
and the addition of a new remote terminal would require a
complex verification procedure to be performed again. As has
already been said, such a disadvantage is inherent in any
command-response system. Another disadvantage is the 5-bit
address field that allows connection of only up to 32 termi-
nals to the multiplex bus.
The 5-bit subaddress field is for the usage of differ-
ent command modes and message controls as required for given
applications.
Bus efficiency varies from 26.6 to 75.3 percent (Ref 3:
App B-6, Figure B-i). Bus utilization is about 33 percent in
the primary mode and about 10 percent in the backup mode
477
W V-7 - - .A W-IT. . . . ...1 To 64 77 7
..- (Ref 49:23); all response times and data rate requirements
are easily satisfied by the one megabit/second bus.
The subsystems' terminals are complicated by the re-
quirement for block message transfer and simultaneous address
decoding on both buses. Such provisions, however, allow more
suitable scheduling of data transmission and a trigger mes-
sage priority.
Th- subsystems check messages for valid sync bits,
,-5 valid Manchester-II data coding, word length, and odd parity,
and report detected errors to the bus controller.
When an error is discovered, the bus controller will
take one or more of the following actions based on data
criticality (Ref 49:24):
- Take no action;J- Reissue the command over the same bus;
- Reissue the command over the alternate bus;
- Initiate special interface tests for the subsystem.
The broadcast mode, although possible, is not used. The
Notice 1 to the MIL-STD-1553B disallows this mode for Air
Force use, because the necessary suppressions of status4-*0
words are not considered reliable enough.
DAIS--Digital Avionics Information System(Ref 3,Ch.6:61-76 and Ref 66:129-136)
Developed as a laboratory program by the Air Force
Avionics Laboratory, AFWAL/WPAFB, Dayton, Ohio, to help de-
78
fine and explore applications of digital avionics and flight
control, the DAIS could be used in a wide class of synchr'-
* - nous multiplexed applications. Although this system has not
-> been used outside of the laboratory environment, the concept
- is interesting as an approach toward viewing the avionics as
a whole, not just as a collection of previous defined sys-
tems. A partitioning of the avionics system was defined in
DAIS such that the mission functions could be partitioned
with a goal to achieve maximum utilization of common hard-
ware, maximum sharing of information and resources, and to
7,,, provide expandable system capability.
DAIS architecture consists of multiple federated proces-
sors that provide communications between the processors and
the other system elements (sensors, weapons, and controls
and displays) through the standardized MIL-STD-1553A multi-
plex data bus system. The inclusion of the MIL-STD-1553A
concept assumes that signals above 400 Hz will be kept off
the bus, while the DIAS designers have oriented the system
toward the repetition rate of less than 128 times per second.
The flexibility achieved by the concept is primarily
due to the use of four classes of functionally standardized
"1core elements" (processors, multiplex information transfer,
controls and displays, and the mission software) that can be
integrated in various mixes and configurations to satisfy
the specific mission requirements.
79
,i DAIS Processors. The number of DAIS processors required
is determined by the mission requirements and the mission
software to be loaded. The laboratory implementation has
only four processors, but this is not the actual limit, as
shown in Figure 23.
PROCESSORS1 PROCESSOR 0 2 PROCESSOR 0 N
BUS CONl- BUS CON- BUCON-TROLLER IN- TROLLER IN- N-TERFACE UNIT TERFACE UNIT AM IT
; IMULTIPL.EX DATA BUS
BUS 8B"
REIITEREOT REMOTE SENoRTERMINAL U 1 TEMMINAL 5 2 - TERMINAL 0 N EQUIPMENT
SENSR SESORCONTROL ANDEQIMENT EQUIPEN - DISPLAY
ILIL EQUIPMENT
Figure 23. PAIS Architecture - Hardware (Ref 87:131)
DAIS Multiplex Bus. The DAIS multiplex system consists
of: bus control interface units (BCIU), one for eath DAIS
processor; remote terminals which interface the subsystems
to the data bus; and the redundant multiplex data buses.
80
-i
% %
k~ ~ The system control procedures define the bus protocol
and error (or failure) management. The BCIU, under control
of the master executive, 'allows bus traffic on only one bus
at a time.
DAIS Controls and Displays. This class of the "core
elements" is an integrated set that provide the capability
for system management by one pilot.
Mission Software. The mission software, as shown in
Figure 24 and described in general sense in the references,
consists of application software, which performs the proces-
sing required for a specific aircraft mission application
'C and subsystems (i.e., navigation, weapon delivery, etc.), and
* the executive software which performs the real time system
control and provides services to the application software.
The executive software consists of the master executive,
local executive, and backup master executive (if required).
The master executive is responsible for the system bus
control and resides in the processor designated to be the
master processor. The local executive is located in each
processor and provides the real time services, including
data read and write, and task control to the application
software. The master executive can also reside in a proces-
sor designated monitor and provide a backup to the master
executive. The executives are table driven to permit flexi-
bility to accommodate various avionic configurations.
81
- -- -. . -- .W.W WL - a. *.~ C. % -. r
The mission software is implemented in the JOVIAL J73/I
higher order language utilizing structured programming tech-
niques, standards, and modular software architecture.
MASTER PROCESSOR REMOTE PROCESSOR MONITOR PROCESSOR
APPLICATIONS APPLICATIONS APPLICATIONSSOFTWARE SOFTWARE SOFTWARE
EXECUTIVE EXECUTIVE EXECUTIVES4DFTWARE SOFTWARE _SOFTWARE
(MASTER A LOCAL) (LOCAL) (MASTER A~ LOCAL)
HARD WARE HARDWARE HARDWARE
BUS CONTROL BUS CONTROL BUS CONTROLINTERFACE INTERFACE INTERFACEUNIT UNIT UNIT
MULTIPLEX DATA BUS
Figure 24. DAIS Architecture - Software (Ref 87:131)
Software is static, but, if necessary, can be reallo-
cated by the reconfiguration procedures provided.
Being a MIL-STD-1553 based systera, the DAIS concept
retains most of already mentioned disadvantages. However,
although under centralized control, this concept provides a
hierarchical system architecture with all benefits of such a
82
.'...configuration. The main feature is the flexibility achieved,
fulfilling the easy "miso-omsin changes require-
ment. The architecture is independent of the specific air-
craft application and, to some extent, has the technology
independence that is desired for any advanced architecture.
DAIS allows both synchronous and asynchronous transac-
-~ tions. Synchronous transactions are according to predefined
schedule (a table lookup) algorithm, while the terminals can
place requests for asynchronous transactions only through
the status words during normal operation.
Mode control and error checking (sync loss, dead termi-
nal and failed bus) are implemented in hardware. The termi-
nals' self-test procedures may be initiated by the master
processor, which is also capable to reconfigure the system
by switching to the remaining element(s) wI'en redundancy is
A provided, or by switchover to a degraded capability mode of
operation.
According to the Honeywell analysis (Ref 49:216), the
DAIS is of moderate integrity, where the centralized, pro-
cessor dependent allocation scheme is considered as the main
- - disadvantage. In addition, while hardware complexity is not
considered high, leaving much of interface responsibility to
DAIS's software makes it rather complex and costly.
GAMS--General Aircraft Multiplex Sstem (Ref 80
* .~ Similar to DAIS, the GAMS is an attempt to define a
83
HiDFl8 15 A STUD OF OFTIMAL COM1PUTER NETWORK ARCHITECTURE FOR 2/3DIGITAL AVIONIC SYSTEMSU AIR FORCE INST OF TECHWRIGHT-PATTERSON AFB OH SHOOL OF ENLI..MFKI
UNCLASSIFIED DEC 83 AIFIT/GE/EE/83D-3 F/ 72E
-- W
* 45
M
,2
11112 1J,4 11.6
%o %
m-a
• ,.
1 (I',-
1~(11.25-1
p-.-'.-~*
* . . . . . . . . . . . .. ' ' '
1
.v, general-purpose, nondedicated multiplex system intended for
~the asynchronous intra-aircraft distribution of a very wide
class of signals, but the concept is based on different
~architectural and control philosophies than those of DAIS.~Proposed as the NADC specification in late 1974, there was
,., no available evidence that the system has been implemented
ii anywhere, except that the A-7E, a carrier based attack
i aircraft, has been used as a base for some GAMS loading
analyses. (The last version of GAMS is known as MIL-E-XXX).
~However, the cocp rpssinteresting network structure
and this is the reason that it is included in this analysis.
Basically, the GAMS offers a distributed architecture
~(Figure 25) wherein the transfer of data is independent of
any single control source.
V..;
^,,N CO NTROLLER
~. -.. genral-purpos, nondedicated multple syte"itnddo
aNFgr 2.GM rchitectural an(otoRpiooh e tha 0tos3o3DIS
.84
Howver thcocep prpoe itrs Inewr stuue
Bas ialyr e. GAMS offrs adsturibued arcitetur
-. 84
MUTPE AL
4' ' ' ' ' '. , . , - . . . ' ' ' . ' " ', . . . .
.. ~ ~ IL C ~ ~ ~ ~ '~~ r'2 . ' J .* *
%* ' In contrast to a command/response system, GAMS might be
called a "request/response" system (Ref 80:2-2), since each
user requests bus access when a data transfer is required,
rather than being commanded to do so bya central controller.
The major elements of the architecture are:
Terminal. In addition to the functions normally per-
formed by a command/response terminal, the GAMS terminal is
also able to participate in a polled system protocol (accept
or reject the bus when offered, place a request for message
service, etc.). The terminals are modular units with func-
tional partitioning (for example, the "B-function" depicted
in Figure 26, page 87, accomplishes all the functions of
message initiation, formation, rec:..ption, and distribution
to user ports). There is a logical limit of up to 32 termi-
nals, imposed by the 5-bit address field in bus allocation
words, as seen in Figure 26.
Multiplex Cable. Having up to four cables, each multi-
plex cable contains two to four bidirectional data buses (the
"data channels") and an unidirectional bus (the "control
* channel") for to the polling function, made by a separate
shielded-twisted pairs in a common cable sheath. Each line
is terminated by its characteristic impedance (71±7 ohms at
1 MHz) and allowed length is up to 300 feet. The projected
data transfer capacity of GAMS is one megabit/second per
Y data channel, so that gross capacity of a full four-times-
85
four-channel cable installation is 16 Mbps. Minimum recom-
mended is dual-cable installation, but terminals are four-
port to support installations with added redundancy or capac-
ity needed. The signal coding is the Manchester-II self-
clocking code. Word synchronization is similar to the
MIL-STD-1553 specification.
Controller. The data channels within each cable are
monitored by separate cable controllers, which offer the
channels to the terminals by a polling process as buses
become available. No response from a terminal is interpreted
as no need for the offered channel. Wait time is 5 data bit
intervals plus the maximum one-way cable delay. Each control-
* ler is independent of all other controllers in the system.
Monitor. A processor dedicated monitor performs the
system health analyses based on BITE data received from the
* terminals. The monitor is used for in-flight or post-flight
fault localization, signalization of the system status and
subsequent system reconfiguration (the monitor cannot shut
down any terminal or controller, neither can the monitor's
failures affect the system operation in any sense).
q Bus access is random. A single controller is respons-
ible for its own buses only.
Message formats are depicted in Figure 26 on the next
page. Both synchronous (periodic) and asynchronous (trigger)
messages transfers are provided.
* 86
LNlsn 2' 1.
I;I-ft.'.
96 ft'
cc CC
al 0
fit a.
-61 1"5 a2
_ Wo
%~.Jft' Ett I
- 0~ oft
I- 0o
870
17). The immediate offer of non-busy data buses to terminals
by a polling process over the control channel is a contin-
uous procedure. Bus access is granted to the polled termi-
nal that has a service request for a message transfer in
queue, and the message transfer proceeds.
Contentions between messages are resolved within the
terminals, that queue messages service requests on priority
and/or first-in-first-out basis.
Messages are transmitted by the source terminal as soon
as a link is established. Acknowledgement of message receipt
is sent by the sink terminal only if an aperiodic message is
sent. Block message transfer is allowed without restriction
on the block length, but only if there is at least one
o additional data bus available in the cable at the beginning
of the block transfer. In this way, there is no possibility
that few terminals seize all available capacity by long
messages transfers.I The distributed architecture, higher data transfer
bandwidth, and emphasized integrity are the primary benefits
that could be gained from the concept. Data load is shared
equally among the system buses that operate independently and
simultaneously. Since this is a normal procedure, there is a
* little reconfiguration needed in case of a bus (or entire
* cable) failure because the load will be again equally redis-
'a 88
tributed among remaining channels.
Although the quadruple controller redundancy can be
considered as a reliability advantage, it introduces more
* hardware and weight in the system at the same time. On the
other hand, the GAMS controller is a relatively simple
device; much simpler than a command/response controller, for
4 example.
Bus utilization and bus efficiency are similar to those
of MIL-STD-1553 systems, depending on the particular system
requirements (number of data words in a message; number of
asynchronous messages that are always acknowledged; command-
response, polled or combined mode used, etc.). Note that bus
allocation messages and data transfer messages are decoup-
led. In that sense, the data bus efficiency is 80 percent in
most cases, which is the theoretical maximum for a command-
response system.
Service requests for periodic messages are originated
at the sink terminals. The use of prediction techniques can
decrease data latency, inherent to any approach where the
* user first asks for data and then waits to continue proces-
sing after reception of the data. However, when a terminal
accepts the offered bus by the polling response message that
commands another terminal to transfer a data (or data block),
then the effect is the same as in any command/response
system, provided that the polling sequence is timed with
989
- * ~ . *47. 77-71%7.7_77 *-077:19
'*0..the commanding terminal's needs. This timing is assured by
the use of clock pulse counters in each terminal and by
update rates for information stored in the terminal's firm-
ware.
There is the possibility that a source is busy sending
* or receiving data over one channel while at the same time it
is requested to send or receive data over the other. In such
a case, the new request is ignored, causing the demanding
terminal to re-request the data. This seems to be the main
disadvantage of the approach.
When a faulty message is received, the terminal decides
whether to leave its service request flag for the same
message, awaiting the next polling offer or to give up the
demand for a retransmission of the source data.
Aperiodic (trigger) messages are originated at the
source terminals and are transferred at the first polling
opportunity. Some delay is introduced by the waiting for the
receiver acknowledgment that it is not busy, but the method
is still more efficient than asynchronous messages handling
in a command/response system (for example, the average de-
lays for aperiodic messages in the GAMS simulation for A-7E
aircraft were about one millisecond or less - Ref 80:2-8).
* Broadcast mode is not disallowed, but is not supported,
as well.
Changing the number or addresses of the terminals will
4 *, 90
* require changing the polling sequences in each controller
and can cause some problems in timing of data transfer.
Software interface is of moderate complexity and is func-
tionally partitioned.
HXDP--Honeywell Experimental Distributed Processor
-V Multiplex System (Ref 49:77-85,185-186)
Intended initially for research purposes with regard to
applications of distributed processing in real-time command
and control systems for small ship guidance, navigation, and
display requirements, the HXDP is also being used in a
weapon control system (Ref 49:77). The system configuration
is shown in Figure 27.
GLOBAL BUS4: -t~........ a-
BUS BUS BUSINTERFACE INTERFACE INTERFACEUNIT NO.0 UNIT NO.1 UNIT NO.N
PROCESSINPROCESSING PROCESSINGELEMENT ELEMENT ELEMENT
UNIT NO.0 UNIT NO.1 UNIT NO.N
Figure 27. HXDP System Configuration (Ref 49:75)
* '. The HXDP bus is shielded twisted pair. The modulation
91
• 0
scheme is Manchester-II encoding. Synchronization is by mes-
sage, not by word, using a 5-bit time sync at the start of
each message (the SOT in Figure 28).
Messages are addressed to software processes ("destina-
tion name" in Figure 28), not to hardware elements. Messages
are variable length, specified by an explicit word count
field, allowing up to 255 16-bit words plus eight bits of
application data in the header (Figure 28). The minimum
message length is zero data words (only the header is sent).
* All transactions are one-to-one, and are always termi-
nated by an EOT signal. The signal is also used as the bus
allocation signal by a specific allocation scheme.
A positive acknowledgment is returned for an accepted
message and a negative acknowledgment with a status word
describing the problems is returned if the message was re-
fused. No response from the sink terminal is assumed as the
negative acknowledgment and retry of the message will be
attempted up to three times, when the EOT will reallocate
the bus.
A minimum length transaction requires about 100 Usec
("zero" data word, properly received), giving bus efficiency
of about 10 percent; a maximum length normal transaction of
255 data words takes about 3.3 msec, which translates in a
bus efficiency ratio of more than 98 percent. The longest
possible transfer (255 data words with no proper reception
92
, until the third attempt) requires about 13.4 msec, bringing
the bus efficiency down to about 33 percent.
--Ow12 BITS- --16 BITS - 5 BITS4*--
z DESTINATION NAME SOT
SOURCETIME SLOT
SOFTWAREDEFINED WORD COUN
255 WORDS MAXIMUM
I ap
I II 1CRC II I
# = RESERVED FOR POSSIBLE BI USEN = CHOICE OF ONEP = PARITYSOT = START OF TRANSACTION (SYNC)CRC = CYCLIC REDUNDANCY CHECK
Figure 28. HXDP Message Format (Ref 49:80)
Bus allocation is decertralized and asynchronous (i.e.
variable length messak. Ae. tndent), according to the "Vector-
93
'i .<, Driven Proportional Access Mechanism", shown in Figure 29.
- - - ac*
-U0 10
0 y 0 vow
mu 3 amINP
, ' i- )smiwa
OW SWAU V SKIIWA49 A SW INTERFACE 04
Figure 29. Vector-Driven Proportional AccessMechanism (Ref 49:186)
The mechanism allocates the bus to only one BIU at a
time. Each BIU is assigned an explicit circular binary
valued allocation vector and an allocation vector index. All
BIU vector indices are incremented at the same time, start-
ing from the same point in the vectors as initialized at
0-,. power-up. The indices increments occur each time at an
94
allocation signal (the EOT) on the bus. The BIU having a
"one" in the position currently pointed to by the index is
allowed to transmit. (No two or more BIUs have a "one" in
.4 the same bit position in their allocation vector tables). A
BIU that has nothing to send at its time, reallocates the
*bus by sending the EOT only.
In fact, the technique uses a polling-based scheme,
where the BIUs are implicitly polled in a predefined order
by the table lookup algorithm resident in each BIU's soft-
*ware. In such a way, the explicit offer through separate
control signals is avoided, thus significant savings in the
message overhead, system weight, and system size are ob-
tained. Also, the system integrity is enhanced in that bus
allocation control signals are kept off the transmission
bus, thus not being overlapped with data transmissions
Vthrough the same bus.
These advantages hide a couple of disadvantages:
*"j (1) Any change in number of BIUs connected to the bus
would require corresponding changes in the vector tables in
all BIUs. This is absolutely necessary if a BIU is added and
may require costly changes in the system's software due to
changes of the polling sequences for some critically timed
messages. On the other hand, if a BIU is deleted, the tables
need not be changed -- the only consequence would be wasting
some time when the deleted BIU's turn comes. Namely, the
. 95
' o , . , 2 o 2 '€ 2' ' . "vv'.-". i; ' .. ' ". .. ,.,.
% missing BIU will be handled by the procedure provided for
the case when a BIU fails to send the EOT to reallocate the
bus: the time-counter with the smallest time constant (each
BIU has such counter with unique time constant) will time
out and generate an EOT reallocating the silent bus.
(2) More serious problems would cause a healthy BIU
which, for any reason, did not "see" the reallocation sig-
nal orderly sent by another BIU. In such a case, the BIU
will not increment its vector index when the others do, and
sooner or later, will try to transmit at the time when some
other BIU transmits having the right to do so. The problem
should be solved by the potentially colliding BIUs: each
BIU, during every message on the bus, normally compares the
allocation information (contained in each EOT) to its inter-
nal allocation data. In such a way, the comparison, plus
history data should help the BIU to determine its own cor-
*rectness and eventually inhibit its own transmitting. There
was no exact data available about the efficiency and relia-
bility of such solution tested in practice, but the software
is clearly complicated due to the complex collision resolving
scheme. In addition, even if the BIU discovers the colli-
sion, a message may be lost (or unacceptably postponed) and,
furthermore, this can happen several times while the BIU is
trying to resolve the problem. In that way, a healthy BIU
with potentially important messages in its queue could be
96
out of the system for a while, just because of a transient
error that caused one EQT to be overlooked. It seems that
such possibility could cancel potential integrity advantages
of the approach.
Loop Multiplex System (Re 49:146-161)
A loop architecture was designed by Honeywell to empha-
size the use of mature or relatively well-understood technol-
ogy and architectural features, while maintaining some
reasonable level of integrity and performance.
The system has up to four loops, all simultaneously ac-
tive. Each loop interface unit (LIU) can be using only one
loop at a time, with the exception that an LIU currently
receiving a message is also capable of sending a busy-
refusal message in reply to an incoming message on any other
loop. The system topology is shown in Figure 30.
All loops are unidirectional, allowing use of point-to-
point fiber optics links. All LIUs participate equally in a
Pierce loop control scheme (Ref 49:151). The Pierce loop is
organized as a collection of fixed-length "packets", or
message blocks, which continuously circulate around the loop
-F through two-packet shift registers buffers in each LIU.
The number of packets in the system is defined by the
product of the number of LIUs and the number of packet buf-
fers (two) per LIU.
97
UNI
LOP0
LOPLO
INEFAEITEFC
II
The circulation of "packets" around the loop is illus-
trated in Figure 31 on the next page. Each packet contains a
* status field which indicates various things about the pack-
~ *:~:.et, including whether it is "empty" or "occupied".
98
, .%. .. ,
LIU 1 BUFFER
LIU 2 BWFER
PACKET N *PACKET M-1
i% ,-.Figure 31. Pierce Loop Organization (Ref 49:152)
,-i An LIU sends a message by first waiting for an empty
~packet to arrive in its buffer and then writing the message
j into the packet as it is shifted out of the buffer. The
packet is shifted around the loop until it gets to the
99
destination LIU, which then copies the message contents,
sets a flag in the status field acknowledging receipt of the
-• message, and allows the packet to be shifted out onto the
loop again. The packet returns around the loop to the source
LIU, which then marks the packet "empty", making it avail-
able again.
Each word in the packet begins with a synchronization
waveform, followed by three bits which specify one of five
sync types, followed by a 16-bit word (Figure 32).
~~u - SYNC - m ___
91 ILLEGAL SYNC IF4 MIANCHESTER TYPE WORDWAVEFORM________ ______
'a.3 BITS 3 BITS 16 BITS
Figure 32. Loop Word Format (Ref 49:156)
Packet format for the loop system is shown in Figure 33
on the next page. As seen from the figure, destination name
is the process name to which this packet is intended. The
word contains 11 bits of name and 5 bits of single-error-
correcting code. The source name is similar. Status/Controla.
bits contains the various flag bits and fields used in the
Icontrol and management of the packet: one bit is for Empty-
.- Occupied, one for Going to Receiver - Coming from Receiver,
q100
i .a . -.. ,. . ... . . .. ... .. '. , . ,., ,., . . ,, , - - - , - - . ,.' , , . . .,
one for Acknowledge/Negative Acknowledge, five bits for
Requested Message ID, and last five bits for Single-Error-
Correcting Code..-.
In the fourth word, eight bits are reserved for the
word count field or the reason for refusal.
SSiP/ DESTINATION ERROR Es EMPTY / OCCUPIEDSOIP NAME CODE 0 - EMPTY
1 a OCCUPIED
Cm SOURCE NAME ERRORCODE Ts TO / FROM RECEIVERCW E T MA Me ERROR 0 - FROM
ID CODE 1 - TO
." L L WORD'COUNT ERROR As ACK / NAKD _ CODE 0 w NAK
DV DATA I I = ACK
DW DATA 2 RC, RETRY COUNT- 00 = ORIGINAL TRANSMISSION
DV DATA 3 O a RETRY IDW DATA 4 10 - RETRY 2- D11 - RETRY 3DV DATA 5
DV DATA 6 LD. LOOP DEMAND. DV DATA 6 _0 a USE ANY LOOP
CV CRC - 0 1 = USE LOOP SPECIFIED BYEOP CRC - LN TO SEND THIS MESSAGE.
LNs LOOP NUMBERIDENTIFICATION OF WHICH LOOP WAS(LD a 0) OR SHOULD BE (LD - 1)
SYNC TYPE, USED FOR THIS MESSAGE.
SSIP - START SOURCE INITIATED PACKET$DIP - START DESTINATION INITIATED PACKETCW a CONTROL WORD
--. DW - DATA WORDEOP a END OF PACKET
--- J
.'-" Figure 33. Packet Format for Loop System (Ref 49:158)
101"p,"
'p'-
~ There can be up to six data words in the message. Space
for all six is always reserved in each packet, but the word
count specifies how many of the words are valid.
Z4 All packets include two CRC words, providing the neces-
sary low-level probability of undetected errors.
The system architecture is designed, as is said, with
the primary goal to exploit "low-risk", well-known technolo-
gies and this is the main advantage of the approach. At the
same time, much of weaknesses and disadvantages of the
technique result from the very same reason.
Bus efficiency is very low - from only 6 percent when
one data word is transferred to the maximum of 36.3 percent
when all six data words are transmitted within a message.
These figures are even less when message retries occur
(there is up to three possible retries after the original
transmission) or if the fact of messages circulation is
encountered. Namely, when a receiver copies the message
content into its buffer, the "IT" and "A" bits are set to "0"
and "1", respectively (Figure 33), but the same message
continues travelling toward the originator which is respon-
sible to "empty, the packet. Such method enhances the system
reliability, but the data throughput is lowered by the fact
that packets are occupied for some extended time after the
message has already been received properly. For the same
reason, message latency could be higher than acceptable for
102
?.~ avionic systems, while already used packets are passing by
andcanot e ueduntil made "empty".
There is no way provided to give any priority to trig-
ger messages. Block messages transfer is not provided. The
broadcast mode, although possible, is not supported.
A loop can be used only if all of LIUs connected to
this loop are healthy, thus jeopardizing the system integrity
by only one faulty LIU. Garbled destination name in a mes-
sage will not be recognized by the receivers, and the packet
may circulate through the loop forever, disallowing the use
of the packet for any other message. Garbled source name has
similar consequence, because the packet will never be made
"lempty".
Passive Bus System (Ref 49:161-178)
The system is also designed by Honeywell in a labora-
tory environment, with integrity (i.e., fault tolerance) as
* the primary goal. Electrically passive radial arm couplers
are used to configure the buses, which are redundant and
used simultaneously by BIUs.
The system contains four passive buses in a redundant
configuration, as shown in Figure 34.
The buses operate concurrently and independently. Each
message carries the identification of the bus selected by a
BIU to transmit the message. This helps the BIUs and subsys-
103
.~ \. ~ tems test for and identify bus faults. Bus failure causes
little reconfiguration: the BIUs simply stop sending and/or
receiving on the faulty bus.
Each BIU is able to transmit on only one bus at a time.
It can receive on all buses at once for the purposes of
detecting a "start of message" sync, looking up the name of
subsequent message, and responding with a busy-refusal, but
it can receive only one message at a time.
Bus BUSBusITERFDAC'E INTERFACE INTERFACEjUNIT UNIT UI
AVIONICS 1AVIONICS 1AVIONICSSUBSYSTEM ISUBSYSTEM jSUBSYSTEM
Figure 34. Passive Bus System Topology (Ref 49:163)
104
~ Global buses are implemented using fiber optic cables
as the transfer medium. Each bus contains a radial arm
coupler (RAC) and several arms, with a bifurcated fiber bun-
dle connected from each terminal to the RAC.
There is a sync on every word of a message, rather than
just one at the beginning of the message. This improves the
* accuracy of the demodulation. The Manchester-II encoding is
used again.
Bus control is decentralized and distributed across all
BIUs. Bus allocation is based on an "Ethernet"-like scheme,
which allows BIUs to transmit on the bus any time it is
I available. Namely, in the event that more than one BIU
starts at the same time, a "collision" is said to have
occurred and they all stop sending immediately. Each then
waits a slightly different interval before attempting to
0' retransmit. The one with highest preassigned priority "times
out" first and gains access to the bus, completing its
message, while all other BIUs wait until the bus is once
again available. Some time is, therefore, wasted while col-
liding BIUs "time out" and it can be serious when demand for
the bus is high and collisions are frequent (Ref 53:310).
Bus allocation is synchronized by the "bus available"
(End of Transmission) signal which is transmitted on the bus
by a BIU as it vacates the bus. Failure to terminate a
transaction properly is detected by a minimum period of bus
105
*~. ~>inactivity. When this occurs, another BIU will uniquely
time out (different time-out constants in each BIU) and send
the EOT to synchronize bus allocation.
All messages are transmitted on a one-to-one basis with
no third party involved. All messages are addressed to
software destinations but, because of integrity goals, both
source and destination names are also included. Each subsys-
tem is allowed up to 31 names which will be recognized by
that subsystem's BIU.
Only one message is transmitted per transaction. Word
* format is the same as for the Loop System (Figure 32, p.97),
and organization of message formats is also similar enough
to be avoided for separate presentation in this subsection.
4 4 The positive acknowledgment message format consists of
two words containing both the destination and source names.
The negative acknowledgment will have, in addition, a BIU
* status word detailing the reason(s) for rejecting the mes-
-. . sage. The EOT signal, also, consists of three words: desti-
nation name, source name, and the bus allocation number.
The 11-bit destination and source name fields permit up
to 2048 global names in the system. Since each BIU can
V recognize up to 31 names (one is reserved for the BIU it-
self), there is enough name space for 64 subsystems. Five
bits of error detection code on each name will detect up to
two single-bit errors. Following the data is a 16-bit CRC
ZI106
-.; T7 .W.'4 Q . -. *-J -j -J .P -7-7
f -. -. code which is required to provide adequate fault detection
over 258 words (there are up to 255 data words mna message).
S Although the organization of message and word formats
are very similar to those for the loop system, the bus
efficiency of the passive bus system is higher in spite of
its greater overhead (explicit acknowledgment words and
three words within each EOT signal). The bus efficiency
maximum value is around 70 percent (a message with 255 data
words, one positive acknowledgment and one EOT). The minimum
'a value for bus efficiency is about 7 percent for only one da-
ta word transferred, and is close to the minimum value for
the bus efficiency of the loop system. The doubled value for
the maximal bus efficiency of the passive bus system is the
consequence of longer messages permitted.
p Block message transfers and/or broadcast mode are not
provided because they were considered potentially more com-
plicated from the BIU hardware complexity viewpoint. Even
such slight increase in hardware complexity was rejected by
the system designers who wanted a system with maximal integ-
rity.
The most valuable conclusion that can be drawn from the
approach (Ref 49:216, Table 2.4) is the fact of a very
reliable collision bus allocation scheme, which allows de-
centralized arbitration about the bus usage. This is made by
the colliding BIUs, without a third part involved. Although
107
Z i. high risk is unavoidable when such mechanisms are to be
developed, the potentially high reliability that can be
achieved encourages further research in the field to inves-
tigate other possible collision schemes.
Active Bus System (Ref 49:179-196)
The system is another (Honeywell) research program,
this time oriented toward achieving high performance (fast
response time and high throughput), but without raising
subsystem software costs.
The "active bus" system is so called because the inter-
connection bus is simulated by point-to-point links between
% adjacent BIUs and electrically active (i.e., powered) re-
peater logic in each BIU. However, the links do not form a
loop. Topology of the system is illustrated in Figure 35.
The avionics subsystems are all attached to a set of
bit-serial global buses via special purpose processors
(BIUs). The global buses operate concurrently and indepen-
dently, and a bus failure causes little reconfiguration: the
BIUs simply stop sending and/or receiving on the faulty bus.
In a fact, each global bus is simulated by point-to-
point links between pairs of BIUs. This avoids the risk,
attenuation and limited number of arms which are presented
in adil rm ouperbuses. To create the effect of a bus,
two unidirectional fiber optic links are used between the
108
-a . - . . -. . . . .... - .
"€ pairs of BIUs to avoid the power loss from bifurcation of a
single link (it is assumed that fiber links which can trans-
mit in both directions at once do not exist). The modula-
tion scheme is again Manchester-IH (BiPhase-Level) encoding.
AVIONICS AVIONICS AVIONICSSUBSYSTEM SUBSYSTEM SUBSYSTEM
BUS BUS BUSINTERFACE INTERFACE * INTERFACEUNIT UNIT UNIT
008
4
U2S
Figure 35. Active Bus Topology (Ref 49:180)
Bus allocation is done in a decentralized manner by all
BIUs. Each bus is independently allocated. The bus alloca-
tion scheme is the "Vector-Driven Proportional Access Mecha-
nism", already explained by Figure 29, page 94.
Both one-to-one and broadcast messages exist in the
protocol. All messages are addressed to software destina-
109
- . 4
- tions. Broadcast messages are explicit (signaled by a bit in
the message).
Only one message is permitted in a transaction, with
positive or negative acknowledgments (the later case inclu-
des a status word detailing the reasons for rejection). The
word and message formats are similar to those already ex-
plained (Figures 32 and 33, p. 100and 101, respectively) and
the illustrations will not be repeated herein. Up to 31 dif-
ferent messages can be defined for each source name in a BIU.
Messages are transmitted from FIFO output queues in
each BIU. The output queue contains pointers to messages,
not the actual messages themselves, which allows rapid inser-
tion, deletion and resequencing of messages on a priority
basis. Receiving is done by subsystem memory buffers. Thereare up to 32 input buffers per subsystem, one for eaci des-
tination name.
The goal of high performance (i.e., bus bandwidth and
efficiency) dictated that the designers wanted to avoid
rejecting messages, thus as much decreasing message transfer
retries as possible. It led to an interesting solution for
the BIU's repeater logic. Namely, if a bit in a message is
received as an illegal Manchester waveform, the message will
not be automatically rejected. A special circuit, called a
modulation correction unit, will arbitrarily convert detec-
ted illegal bit to a legal bit waveform. Obviously, with the
110
S X--
7.- -. 3
- .-.. 50 percent chance the bit will be correctly restored. if
not, there is high probability that the error will be detec-
ted and corrected by the use of the message CRC word. The
net result is that the message has not been repeated and bus
efficiency as well as data throughput is enhanced.
The implicit polling by the use of the Vector-Driven
- -' Proportional Access Mechanism (Figure 29, p.94) introduces
possible weaknesses that have already been discussed. In
addition, the mechanism has the disadvantage in that it
depends on the subsystem transmission needs being accurately
estimated a priori (at system configuration time). This has
also been discussed earlier (page 56).
In normal operation, bus efficiency is similar to that
of the passive bus system due to the same message and word
formats. However, both the message retries avoiding scheme
in the active bus system and the fact that direct collision
between messages does not occur in this approach (thus
eliminating the colliding time-out of the passive bus) make
the active bus system more efficient in the view of data
* throughput.
Switched Network (Ref 50)
The switched network approach evolved from investiga-
tions at General Dynamics/Fort Worth Division. In essence,
the technique works very much like the public telephone
A system, where physical connections are established between
devices only when a message is to be transmitted and only
for the duration of the message. The basic hardware to
accomplish switched network communications has been demon-
strated in the laboratory, while the protocol is currently
being implemented in hardware and firmware, and a demonstra-
tion is scheduled late this year. The basic topology is as
shown in Figure 36 (page 113).
A key constituent of a switched system is the switch-
ing element (center). Each is implemented with a number of
individual ports, one for each attached intelligent termi-
nal. Switching commands are provided by the terminals. A
switching center port may be connected to a terminal or to a
port on another switching center. In either case, it re-
sponds to commands directing it to tie its lines to those of
another port in the same center. By external wiring, the
same switching center port can therefore support a terminal
directly or act as a gateway for a long distance connection
between switching centers.
The architecture is considered to have ability to use
fiber optic media if necessary, avoiding the splitting of
energy problems by the point-to-point communications pro-
vided. In addition, each port is connected to its terminal
with a full duplex line, which allows each connection to be
tested before data is transmitted.
112
g.
I.o
r7,7
TOTAL SYSTEM LEVEL COMPUTING PlOIPDJWAE E TOSAIE
- \' GLOBAL MULX NET- DUAL SWITCHING CENTERS Hw
-- , -- ------ -
xa x
REGIONAL MUX NET -DUAL SWITCHING CENTERS /
-I -- - - - -- ---
10C&. COUUTJN.0
"-.LOCAL MUX MET- DUAL SWITCHING CENTERS MaP336M
:~PA F11000 wpmN-- -
EH P33MW M N
Figure 36. Switching Network Architecture (Ref 50)
The technique asmsthe use of "smart" terminals
Th assumes
'.4' (controllers) which have direct memory access to the memory
of the main computer and which can act as both a passive
terminal or active controller. Only one terminal is needed
113
. ,. . , . . ..- ..- . -
a T . . . . . .b . . .
°,.
i "per processing element to provide communications with any
other element in the system, anywhere in the hierarchy, but
systems which have a great deal of communications would be.5
normally placed on the same switching center.
Any terminal wanting an information transfer first
issues switching command to establish the link, and then
transmits data as indicated in Figure 37. A message identi-
fication word always precedes each message to identify the
message content. Parity is transmitted with each word.
TRANOMIT - TeUT A DATA KM TO A NI.UTU TUM.
__ CONTROLLER ACTION Fe M--solu llol vm tll E l I
* TERNINAL Wm8OiBEi C STATUS
RS R - IIOT A DATA LOM FMK A 8n.WE= TMWDL
SCONTROLL ACTION I I mb
TRANSMIT VITH 1C140 - TIAWifTO A DATA SOW 34104 I94UVU RI.
0 CONTROLLER ACTION re =O2 0001WRD*
Figure 37. Switched Network Message Types (Ref 50)
Initiation of messages is controlled by command tables
in the memory of the processor initiating the transaction,
_ with a series of consecutive commands constituting a command
4 "114
4'."4- .'. ,,;.,, ,,...,-',..c ,[% ,, ., ".. .. • . :.-% . "• , . . : . . "- .. ". . - ._. , -- .- " . .
table. Each terminal, once initiated by its associated com-
puter, will process all the commands in the command table
before requiring further support from the computer. Figure
38 shows the format of each command in the command table.
0 1 2 3 4 5 6 7 B 9 10 11 12 13 14 15
OPCODE OPTIONS WORD 0S1 cl Al 81 RT I E I W1lW21W3 D
MESSAGE ID j WORD 1
-DATAPOINTER WORD 2
NUMBER OF DEVICES TO BE CONNECTEDJ WORD 3
DEVICE ID WORD 4
DEVICE ID j WORD N
Figure 38. Switched Network CommandTable Format (Ref 50)
The first word (word 0) indicates whether the command
is a transmit, a receive, a "broadcast" call etc. (including
video and video broadcast commands), causing the switchingcenters to switch video as well as digital lines. Maintain-
ing the connection to a subsystem after the completion of a
message is also provided as a command by this word, as well
as a Selectable Wait specifying how long to wait for a busy
terminal to become free before aborting the message.
.. The third word is a pointer to the location of the ac-
115
gig
-- S - - -- - - 1
tuldata to be snwhile the fourth and subsequent words
in each command table are used as pointers to routing ta-
4. bles, which contain information used at the switching level
in completing the desired connection(s).
The protocol requires each switched network terminal to
support only one command table at a time, but other messages
may be received simultaneously on the second switched net-
work channel.
-~ The approach is highly modular in nature, so that sys-
tems can be constructed from standard modules. This is the
.4main advantage of the technique, because the modular struc-
'S ture may enhance the system stability, improve interbus
* communication, and eliminate the lack of easy growth capa-
bility found in some classical bus oriented systems. On the
other hand, the considerably more hardware and wiring requi-
red can decrease overall system integrity and increase total
size and weight of the systems.
The approach supports block and string message trans-
fers, as well as trigger message transactions. The message
I. latency could be decreased by placing subsystems with high
ratio of communications to the same switching centers or to
'S the centers that are close enough. However, the overhead
imposed by the protocol naming all centers to be switched
'5' and searching for available paths, can make this latency
unacceptable if longer paths are necessary for any reason.
'5- 4.-,116
*,b.:*"-:".. Bus efficiency, as defined for the purpose of this
study, does not have the same meaning for the switched net-
work approach as has for all other techniques described in
this chapter. Namely, for the same amount of data transferred,
bus efficiency in the switched network considerably depends
on the overhead imposed by naming of all switching centers
' involved in the transaction. However, to have at least a
rough feeling about the bus efficiency, it is calculated to
Is be 25 percent for only one data word transferred through a
center where both the source and receiver are connected.
There was no data available about the signal modulation
1. code and synchronization pattern used. From the fact that no
clock lines are shown in Figure 36, p.113, it can be concluded
0 that one of the self-clocking modulation schemes is used.
The use of the mass memory approach can be critical in
the case of heavy load in the network.
The technique is still under development, so that data
about software complexity was not available when this report
was written. The software can be expected to be fairly com-
plex, as much as interface software in any switched matrix
technology.
I-
DP/M--Distributed Processor/MemorySystem (Ref 49:68-76)Y
DP/M system was designed for the computational require-
ments of air-to-ground attack and electronic warfare mis-
I1 1" I' 117
S'.
4.'- ..I-',,,' . ,.,.- .. wL . ' : .' .,.. ' ' .. ' ._' " .. ' .. - . .. ' ' ,.,..- ' ' " " , ,-','' ... . . ._: "- . .
sions on an advanced aircraft. The initial design was done
by the Honeywell, but revision and more detailed design was
done by the Texas Instruments.
4
PE-1 PE 2 P 3PE 4 P
L LOCAL BUS A LOCAL BUS B
Figure 39. Texas Instruments DP/MY.F Configuration (Ref 49:68)
The structure consists of a single (Honeywell) or dual
redundant (Texas Instruments) global bus connecting all of
several local buses and processors (Figures 39 through 41).
The global buses carry low rate and system-wide mes-
sages, while local buses provide the higher data rate paths.
The local buses are switched in the Honeywell system and
fixed in the TI approach.
The total data rate requirements were about 500 Kbps to
760 Kbps, and speed was one Mbps raw bit rate. The medium is
.9 twisted shielded pair wire with maximum length of 300 feet.
118
° , *..* i .. *~-- ~ . . .
,u,
BUS INTERFACE
.,., PROCESSOR F
POINT- TO - POINT"* (SWITCHED)
BUSES
Figure 40. Honeywell D/M HybridBus Topology (Ref 49:69)
Modulation is Manchester-II and synchronization is by
message. In the TI technique a "header sync" follows the
header word in each message, delimiting the header from the
message text.
One-to-one and broadcast transactions are supported
(only broadcast in the TI solution). Following the last data
word is a bus sync indicating the end of message and reallo-
cation time. There is no response from the receivers.
Message lengths are from zero to 512 data words in the
Honeywell, and up to eight data words in the TI approach.
119
S.i
4LOSAL SVITCHED BVITCHIED SVITCHED SWITCHED-6
BUS A S C D
*6
5%%
SWITCHOPEN
+ CLOSED
Figure 41. Honeywell DP/M SwitchedLocal Buses (Ref 49:69)
In the Honeywell solution there is no header; the mes-
sage begins with a variable length name which is recognized
by four software-controlled registers in the BIU. In the TI
solution, a 1024-bit table in the memory of the processing
elements serves as an indicator which message ID numbers are
to be accepted.
The global and local buses are controlled and allocated
identically. The local bus has its own control hardware, se-
parate but equal to the global bus hardware. The bus is re-
allocated at message length-dependent intervals. The Honey-
.: Z 120
- - - S ' . . . -6 . - . - - . .. o - 5 .- .- .- .
, .well explores their Vector-Driven- Proportional Access Mecha-
nism (Figure 29,p.94), while TI utilizes a prioritized round
robin scheme: each processing element gets control of the
bus once in every "n" time slots, where "n" may differ for-q
some processing elements, allowing prioritized allocations.
In fact, both the control schemes are implemented in a
similar manner, using implicit polling techniques. The BIUs
"know" their turn by counting bus allocation time slots (in
the TI approach) or by the table lookup algorithm with
pointer increments at each EOT (in the Honeywell approach).
The primary advantage of these distributed control approa-
ches is that control signals are decoupled from data transfer
paths. The primary disadvantages have been already discus-
sed: implicit polling technique requests that subsystems
data transfer rates are known and set in advance, thus
introducing inflexibility when subsequent changes of the
system are needed. In addition, problems arise when a BIU
overlooks its turn (or sees a nonexistent EOT, which is not
likely but is still possible). As has been discussed (pages
i 95-97), such problems require special handling, thus compli-
cating software and decreasing system integrity to some
Nextent.
The system integrity is further lowered by the fact
that there is no response from the receiver(s). The acknow-
ledgments are essential for system integrity and the message
• './. 121
*44"'-. - ,' ' .- ,- ..-. .- . , . .. ,,.<. .. ,. ,'.S'.,'.,',.
-7 "
- "-." sender should know what happened to the message during the
transfer (at least for some classes of messages). The situa-
*- tion is even more aggravated by the Honeywell solution for
receivers that react on any invalid message by simply igno-
ring it. The TI terminals ignore messages with invalid
headers, while garbled data can be used under some circum-
stances.
The local buses are not backup for the case of a global
bus failure. Only one global bus can be active at a time (in
the TI system there are two redundant global buses, while
Honeywell's switched busses are backups for the global bus).
- The backup bus switchover procedure is also decentralized
*and resident in each processor's memory.
There was not enough data to calculate bus efficiency
A4 for either system, but considerably longer messages permit-
ted in the Honeywell solution lead to the assumption of its
4' higher bus efficiency percentage.
Transparent Contention System (Ref 53, 55, and 61)
The Air Force Wright Aeronautical Laboratories (AFWAL)
at the Wright-Patterson AFB, Dayton, Ohio, has recently
demonstrated the Continuously Reconfiguring Multi-Micropro-
cessor Flight Control System (CRMmFCS). The system utilizes
originally developed concepts of "transparent contention",
"virtual memory", and "volunteering for tasks" in a contin-.. _
122
.-
'v. "- --.- .".'. "-'. ..- , " , " " " .4,,, -,, .' .*%* - "4. . ...- * - -, , .. -- -. 4 .
- -. , -F -7177%;77 77 7. 747 .
1 ued research program.
The essential elements of the bus architecture are
shown in Figure 42.
CLOCK
T DT
'.4 srialFigure 42. Transparent Contention Conceptual
9. Topology (Ref 55: 30)
The processing elements are interconnected by a common
seralbus made up of a data line and a clock line. Each
processing module consists of a microcomputer with a broad-
4q caster (B) and a receiver (R). The box labeled "T" in Figure
42 is a bus termination circuit which generates the clock
9. signal and terminates the bus properly.
While the use of fiber optic links has been considered,
the actual medium in the laboratory model is wired-OR. The
modulation technique is unipolar NRZ code. Synchronization
is by a separate clock signal. The bus allocation signal is
9 consecutive one's, transmitted each major frame. The data
123
word is 37 bits long and starts with a 5-bit sender ID,
followed by 11-bit variable name, then followed by 16 bitsN of data. The data word is divided into four 8-bit bytes,
separated by zero's to avoid confusion with the bus alloca-
tion signal. All transactions are broadcast only.
The processor wishing to transmit places its informa-
tion in its local broadcaster FIFO buffer. If the bus is
available, the message is transferred starting from the next
clock pulse, and any other processor is prevented (using a
logical "busy" signal) from initiating a broadcast and over-
lapping the message being transmitted.
Transparent Contention Concept. The problem of conten-
E~3 tion, which occurs when more than one transmitter attemptsto broadcast starting at exactly the same clock pulse, is
resolved by a transmitter-bus interface circuit (Figure 43).
OPEN- OPEN-TRANSMITTER COLLECTOR TRANSMITTER COLLECTOR
ATRANSISTOR B TRANSISTOR
-4?SL Figure 43. Transmitter-Bus Interface (Ref 55:32)
124
When any transmitter sends a "zero", it turns on its
output transistor shorting the bus to ground. To transmit a
"one" the transistor is turned off, allowing the bus to be
pulled high by the pull-up resistor if there is no other
transmitter(s) shorting the bus to ground at the same time.
Every transmitter constantly compares what it is trying
to put on the bus with what is actually there. In the event
of disagreement, the transmitter simply stops sending and
wait for the bus to become available again. The disagreement
is noticed only by the transmitter(s) trying to put a "oe
on a bus. Those puting "r" on the bus are not aware of
-~ any competition. The net result is that logic zeros have an
inherent priority on the bus, as shown in Figure 44.
TRANSMITTER 1DESIRED OUTPUT 0OF 1 0 0j 1 0TRANSMITTER 2DESIRED OUTPUT 0 1 1 0 1 1
1 S_
ACTUAL OUTPUTAPPEARING ON BUS 10 11 0 0
TIME (MICROSECONDS) 1 2 3 4 5 6
Figure 44. Bus Contention Arbitration (Ref 55:34)
125
The competition illustrated at the third interval shown
in Figure 44 will actually cause transmitter 2 to stop
transmitting because it is only one that noticed disagree-
ment. As shown in the Figure, at the end of the competition
only one transmitter "wins", completing its message. The
concept works equally well for any number of transmitters in
contention.
dFigure 45 shows the transmitter interface of a single
processor in a system with four buses.
T "CLOCK BUSSES
I I-
AVAILABLE 4 TO 1 4 TO I I TO It' •Dklr~eORSELECTOR SELECTOR DISBTR IBUTOR
I': IABORT ABORT -BIT
2.SH-FT ENABLE C0. COMPARET
TRANSISSION LOIFDC.. ONTROL LOOIl¢61
4CPU
START 7
Figure 45. Transmitter Circuit (Ref 55:37)
126
-4%v
".P . '
~, In the laboratory environment, six such processing mo-
dules were connected to the four buses to demonstrate feas-
ibility of the concept.
The circuit is controlled by the "transmission control
logic". Upon receiving a "START" signal (from the CPU), this
logic instructs the "bus finder" to search for a free bus.
When it finds one, it locks two data selectors and a data
distributor onto the bus (using its "bus select" lines) and
signals the control logic that the bus has been found. The
control logic then shifts data out through the distributor.
A one-bit comparator monitors for possible disagreements,
when an "abort" signal is generated, meaning that the compe-
tition has been lost and that searching for another bus
should be started.
A two-page buffer (P1 and P2, in Figure 46) is memory-
mapped to the local CPU. These pages alternate functions, so
that while one of them is being loaded by the CPU, the other
is being unloaded onto the bus.
Virtual Memory Technique. Within the laboratory pro-
gram, an another concept is also implemented. The technique
is illustrated in Figures 46 and 47.
In the so-called "virtual memory' design of Figure 46
each processor is given its own copy of the "State Informa-
tion Memory" (SIM), containing all information about all
global variables in the system. In that way each processor
127
L.-_P
" ",*- broadcasts data as soon as the data have been computed, such
that all variables are updated in the SIM of each processor.
NOn the other hand, any processor reads data from its own up-
dated SIM, without wasting time waiting for the data to be
sent if needed. In fact, each processor behaves as if it had
access to a common memory.
* nnI .2N !N IZ~ emI 1 PIN
uP uP uP uP uP uPLa0m. LOCAL 1.0M L0L 0C. A0M
Figure 46. Implementation of Virtual Memory (Ref 55:44)
While the transmitter (XMIT) is explained by reference
to Figure 45 (page 126), the receiver and its relationship
to the SIM is shown in Figure 47.
The receiver consists of a serial to parallel convert-
ing shift register (SIPO) and a RAM which is marked as SIMU'.
in the figure. The SIM is mapped into the microcomputer's
address space as a block of read only memory, but is acces-
128
-%.7- 71
sed by the SIPO outputs as a block of "write only" memory.
SERIAL- IN PARALLEL- OUT SHIFT REGITER (SIPO)DA A BUS 0 B TE--O B TE 3 0 BYTE._ _. 2 10 BYTE 1 01
16 BIT INFORMATON
""SIll DATA 4MEMORY UTO CPUWORD ,2k BY 16 BIT RAMT
16BI INORAINSN )RS$(RMCU
V
11 BITE SIMl ADDRESS (FROM CPU)
'II
Figure 47. Receiver Block Diagram (Ref 55:46)
Figure 47 is expected to be self-explanatory one, but
Vnote that there are "n" SIPO shift registers (where "n" is
number of buses) per processing module.
Current high speed RAMs can handle as many as four
accesses per microsecond. Since each word appears in the
SIPO, in the worst case each 46 microseconds (9 bits of bus
-- allocation time signal plus the 37-bit word), thus the only
129
, . ...
-4 ' , ,, , , ,, , . . , . , . . , , , . . , , . . , . ,. , . . ,. . , . . . . .. . , , . . . . , , .t"i,,, ' V ', '''" . '''' . . . '. . '' . - . - . - . "• . " . - , " . ' . " - " ,,." . . " . " .
limitation are technologically feasible connections of up to
8 buses to the SIM.
.4Software Partitioning. The virtual memory concept re-
quires transmissions to be scheduled carefully. The schedu-
ling technique is based on the known concept of "quantized
software". Namely, time on the bus is divided into equal
time,-slots that are one millisecond in duration for this
system. Having messages of 46 bits in length with each bit
taking one microsecond, it is possible for the system to
transmit 1000 +46 = 21 complete messages over a bus during
each time slot. Thus, the system's four buses are able to
transfer up to 84 messages during each time slot. The trans-
*.1 parent contention process packs all messages one after ano-
ther, distributing load equally between the four buses. When
all messages are transferred, the buses sit idle until the
next time slot starts.
The system's software is modularized in chunks of up to
one millisecond in duration (so-called "millimodules") and
at the beginning of a time slot the variables being proces-
* sed during the preceding slot are ready for transfer. This
software partitioning is based on the execution times, rath-
er than on functions performed. Because of the difficulties
to adjust processing chunks to exactly one millisecond
frames, some "mill imodules " are shorter than one millisecond.
It clearly results in wasting time and in an additional,
130
%J1.
T.- VW
8 [-although not significant increase in message latency, but
note that the method allows exceptional efficiency in task
scheduling. Namely, it is now possible to know exactly in
millisecond when a variable will be ready for the use by
- .. other software tasks and when the tasks can be scheduled. In
most applications, such feature can make software design and
testing a lot easier and is a well justified trade-off for
some greater message latency.
The six-processors, four-buses laboratory implementa-
tion is designed with only 22 out of 84 possible messages
scheduled for transfer during each time slot. In that way,
V' the system integrity is enhanced because one half of the
existing four buses can fail, but the system will still be
operational, without any reconfiguration procedures requried.
Although the authors claim 100 percent bus utilization
and efficiency achieved, the actual bus utilization and bus
efficiency (as defined for this report) are only 25.3 and
34.8 percent, respectively. This is a consequence of the
millimodule partitioning method and significant overhead in
each of the 37-bit word messages plus the nine bits of the
4.4 bus allocation signal. Note, hwoever, that the bus efficien-
cy is better than efficiency of any other approach described
r in this chapter, when compared on one data word message
basis.
Message latency is less than one millisecond.
131
4.%
-> Task Volunteering. In such a system, there are always
more processors than global tasks to be performed, due to
the redundancy requirements. This fact led to another major
outgrowth of this research program: the development of the
autonomous control method for dynamically distributing tasks
among a group of processors without using a central control-
ler. The method is defined as "a scheme whereby each proces-
sor independently determines its own next task based upon
the current state of the system" (Ref 55:68). Software
( modularization, as discussed earlier in this section, allows
* that all tasks are precisely scheduled and documented in the
so-called "task assignment charts". The same chart exists
in each processor' s memory and the processors know which
0 tasks should be executed within any time frame. A process,
called "processor volunteering", is used to distribute tasks
between processors during every reconfiguration cycle. Each
processor performs a pre-programmed self-testing procedure,
and if it finds itself healthy, will "volunteer" for a new
task by placing a "1" in the volunteer status table at' the
place corresponding to its identification number. Otherwise,
the processor is assumed to have not volunteered.
A healthy processor will examine the table and assume
that the processors with smaller ID numbers and which have
volunteered will take tasks in the order that the tasks are
numbered. Thus, the processor takes the next task for itself.
-. 132
4IV
~: *~*As an example: processor #5 will take the task #5 if all of
processors #1 through #4 volunteered, but if two of them did
not, it will take the task #3 and so on). Tasks are ordered
either in a priority order (when task counting starts always
from the first task), or are looped and the starting point
is random but known to all processors (in the laboratory
implementation the value of a certain variable with high
changing rate is used as the marker to the starting point).
In that way, reconfiguring the system from time to time is a
normal procedure, rather than an emergency.
The net result is that faulty processors are out of the
system almost immediately after failures occur. Furthermore,
even in the case that all processors are "healthy", due to
the randomized starting point a processor performs different
Stasks after each reconfiguration cycle. This enhances the
system reliability. Namely, if a processor is for any reason
unable to recognize that it has failed and volunteers, it
will probably perform some different task than the task of
the previous cycle. Thus, errors will not accumulate in the
value of only one variable, because the value of erroneous
-'p variable will be refreshed by a healthy processor within the
next time frame. In addition, other processors can shut down
the faulty one if there exists a consensus about which
processor was volunteering through several cycles although
it was not allowed.
S. 133
There are indisputable benefits that can be gained from
the approach and this is the reason that more attention is
paid in this chapter to the transparent contention technique
than to any other. It seems, namely, that the approach has
been successfully integrated in a system that gathers most
of its advantages from inherent characteristics of a fully
distributed technology.
However, some drawbacks are also recognizable:
(1) The need for the clock wires and bus termination
circuits introduces more hardware, thus more weight and less
reliability for the bus itself;
(2) Due to transistor leakage current, open-collector
transistor direct connection to the bus limits the number of
the subsystems which can be connected to one bus. Also, it
introduces a possibility that any shorted bus-buffers may
"kill" the bus. The buffers are considered as part of the
bus, so that failure of a buffer is the same as failure of
the bus, which clearly means many more single points that
decrease the bus reliability;
(3) The effect of propagation delays may limit line
length to the extent that it prevents implementation of the
concept at the global system level;
(4) And finally (but perhaps most importantly), a sen-
der never knows what has really been received and stored in
the receivers memories (SIMs). Transient errors in a noisy
134
aircraft environment may cause an error propagation over
the entire system: a bad data stored in the SIM at only
one receiver can be the cause of miscalculation of several
variables to be transmitted and then received in all other
SIMs, causing several new sources with bad data and the
error propagates. The solution proposed by the system de-
signers is triplication of each variable in the SIM and a
variable must be "voted" before it can be used for the next
stage of a calculation. It introduces more software comple-
* xity and can also mean a greater message latency.
Frequency Multiplex Systems (Ref 22:63-73)
The use of frequency multiplexing techniques to distri-
bute information over coaxial cables offers an advantage
over time multiplexing. The technique is compatible with all
signal types and simultaneously provides continuous outputs
to all subsystems.
Frequency multiplexing techniques are widely used in
cable television systems; however, no single frequency mul-
tiplex bus configuration with general applicability to avio-
nics equipment has been identified. Namely, there are a
number of constraints which affect the utilization of fre-
quency multiplexing for transmission of signals in aircraft.
An ideal frequency multiplex bus for avionics would
allow direct exchange of information between any of the
135
-o'
L.' .['[. subsystems on the bus regardless of the equipment position
on the bus, accommodating multiple channels over a broad
bandwidth.
'This implies the use of directional couplers to inject-
extract useful information onto/from the bus, and the util-
-- ization of amplifiers (which are nonreciprocal devices) in
order to overcome the losses due to the couplers and the
- lossy transmission line, introducing significant limitations
on the overall capability of the configuration by the posi-
tion of the directional devices on the bus (Figure 48).
-'I4.
T1 R1 R2 R3 T2 R4 R5
-., INFORMATION EXCHANGERI R2 R3 R4RS LEGEND Y =YES
T1 Y N Y N Y= NOT2 N Y N N N Ti = TRANSMITTER
Ri = RECEIVER
". Figure 48. Directional Frequency Multiplex Bus--Position Dependent Configuration (Ref 22:65)
A transmitter can send information to any number of re-
ceivers downstream, but a receiver upstream from the trans-
', 136
'(S
,. -..,4
mitter cannot receive the signal, limiting the ability to
exchange information by the initial configuration of the bus
(the use of in-line amplifiers in the position dependent
configuration is limited and once incorporated further re-
duces bus flexibility).
Commercial systems (Figure 49) overcome the configura-
tion limitations by having all the transmitters send signals
upstream to the "head end" where they are converted to a
different channel and retransmitted downstream for receipt;
the requirements for in-line signal amplification are met by
frequency dependent bidirectional amplifiers (low and high
pass amplifiers in opposite directions).
Ti R1 R2 R3 T2 R4 R5
INFORMATION EXCHANGERi R2 R3 R4I R5 LEGEND Y = YES
TIY Y Y Y Y N =NOT2Y Y Y Y Y Ti =TRANSMITTER
R! = RECEIVER
Figure 49. Directional Frequency Multiplex Bus--Head End Configuration (Ref 22:65)
137
Thus, the bus is in reality two buses operating in
different directions and frequency bands on one cable with
centralized control.
Neither the position dependent nor the "head end" bus
configuration is applicable to future avionics installation.
The position dependent configuration does not offer growth
capability and may require excessive cabling to attach the
subsystems to their correct locations on the bus. The "head
* .. end" approach requires an excessive hardware investment,
occupying the two frequency channels for each single trans-
mission.
Dual direction coupler configuration capable of injec-
ting or extracting signals on/from the bus in both direc-
tions simultaneously is realizable, but losses would be
approximately 6dB greater than for a single directional
9~~* coupler (Ref 22:64). This further limits bus length if in-
line amplifiers are not to be implemented.
* Figure 50 illustrates a configuration proposed by the
NADC, which could be applicable. In such two-cable configu-
ration, in-line amplification is feasible and the location
of the amplifiers will not limit the bus operation. However,
the couplers appear complex, although they exhibit 3dB (as
opposes to 6dB) additional insertion/extraction loss for a
given value through insertion loss (Ref 22:66).
'.9 138
.J.
°.-1 R2 R3T5R
INFORMATION EXCHANGE INFORMATION EXCHANGE
UPPER CABLE LOVER CABLE
V R1R2 R3 R4R5 R R2 R3 R4R5 LEGEND YaTYESTI Y Y T T1 NN N N N- N aNOT2I NNY Y T21IY T YN N T1 - TRANSMIITTER
R I a RECEIVER
Figure 50. Two-Cable Frequency MultiplexBus (Ref 22:67)
The development for such implementation would include
development of bus frequency conversion modules, bus coup-
lers (transmit, receive, and transmit/receive), in-line
* amplifiers, and filter assemblies. There was no available
evidence that such development has been initiated, and the
technique is presented in this report only in order to show
a possible alternative to time division multiplex systems
with some of the advantages of frequency multiplexing tech-
niques. Some disadvantages and problems to be solved are
presented, as well.
* iThis is, probably, the right place to mention that
there is relatively new technique which can be considered
S ""139
CC.
-..,=.
-. ~ .>.~.within the class of frequency multiplex systems. Namely, the
optical wavelength division multiplexing (WDM) systems, with
signals transmitted on different wavelengths through a sin-
gle fiber, can have increased information capacity and fault
isolation properties over single wavelength optical systems.
The technique is not more elaborated in this report due to
several technological problems that still remain unsolved at
present, but the interested reader may refer to Ref 82.
ARINC 575 Digital Air Data Multiplex
System (Re 41:48-51)
* Digital data buses are utilized in today's commercial
aircraft, but only on an individual system basis. Wider
) usage is anticipated in the next generation (Ref 26).
0 The current commercial standard (the ARINC-575 specifi-
cation) was written in 1971 by Aeronautical Radio, Inc.
(ARINC), sponsoring the Airline Electronic Engineering Com-
mittee (AEEC) that formulates standards in the form of
characteristics and specifications for airline electronic
equipment and systems.
* The ARINC 575 bus operates only in a broadcast mode. A
single transmitter sends data at a 11 Kbits/sec rate simul-
taneously to a maximum of 20 receivers connected to the same
bus. Bipolar RZ modulation is employed. There is only one
i transmitter on each single bus and if a receiver needs data
from more than one source, it has to have a separate BIU
* 140
~ ~-.:-connected to the separate buses. There is no acknowledgment
sent by the receiver to the sender. Any garbled messages
will be lost since the receivers cannot request a repeat,
nor can the transmitter ever know what has been received by
the receivers.
The message contains a header that names the variable
being sent, followed by the data and any sign or status
indications. A message is always one 32-bit word.
The bus is transmitter controlled. There is no possibi-
lity for any contentions to occur on any bus since only one
transmitter transmits on each single bus, thus bus allocati-
on signals are not needed. Another advantage of the techni-
que is that the transmitter sends a data word when the word
is available to be sent, which is very desirable feature for
real-time avionic systems. However, such good message laten-
cy characteristics are paid for by oversizing of the system
cabling and by multiplicating the BIU's inputs. The problem
is further aggravated if redundancy is required.
As seen, more suitable solutions for military aircraft
do exist, and this approach is briefly presented only as an
I..; illustration of existing status in commercial aircraft.
Chapter Summary
In this chapter, fourteen different approaches to the
same problem of data multiplexing in avionic systems are
141
9.'Pee
W, W k)7F- 71-
-. given. The solutions are described and analyzed according to
the multiplex system key issues defined in Chapter III.
The techniques represent the current system design base
and the advantages of particular technologies are included
in the definition of "optimal" architecture characteristics
in the next chapter.
.14
~ .~>-.V. Optimal Architecture Definition and Conceptual Design
Having defined the networks' figures of merit in Chap-
ter III, and discussed the advantages and disadvantages of
different technologies in Chapter IV, it is now possible to
* define the characteristics of an "optimal" architecture for
digital avionics systems, as follows:
(1) The architecture should reflect the results of the
successful current system design base, i.e., should involve
as many features of the existing time-division serial multi-
plex bus techniques as possible to in~tegrate into a single
architecture.
* (2) The architecture should be simple and general enough
to be standardized for use in the entire aerospace industry.
The very same architectural philosophy should be utilized
from the system's global bus down to the lowest subsystem
level. This will allow a massive, cheap acquisition of the
basic components that can be integrated in various ways to
build a system as required. In this way, the main goal of
any standardization will be achieved.
(3) The system designer must be free to utilize the
philosophy by using any current technology, according to the
particular system requirements. This translates into the
technology independence of the "optimal" architecture philo-
sophy, avoiding the trap of technology dependent hardware
Ism,
143
standardization that can freeze possibilities for improve-
ments as technology advances.
(4) Multilevel, hierarchfical, lossy configuration of
system topology provides possibilities for functional parti-
tioning, both in hardware and software. This will allow
desirable flexibility in multi-mission applications of the
same system, including the possibility for subsequent growth
of the system. The maximum fault isolation between subsys-
tems will also be achieved in such a way.
(5) Distributed bus control with a purely transparent
contention bus allocation scheme is required to support the
flexibility and integrity features of the system topology.
(6) The system must be independent of physical charac-
teristics of the transmission media.
If wire cables are utilized, the transformer coup-
ling is preferred.
(7) Bus length of up to 500 feet is considered to be
enough for any application in the foreseeable future.
(8) Logical limitations imposed by the system protocol
and message formats on the bus size (the number of terminals
allowed), must not be greater than those imposed by the
utilized transfer medium physical limitations. Doubling the
number of 32 terminals (this number is common today as the
limitation on bus size), seems to be enough for the systems
in the years to come.
144
V.
- -. &: - S~ . .S S--. . ' o . -i- -
. .. j . . - -. , h "- -,- .oj . .
'- (9) The architecture is to provide transmissions with
up to 10 Mbps data rate. Bus efficiency and message latency
should be improved as much as possible by decreasing the
message overhead. The information-only transfer philosophy
will support these improvements, bringing up the new quality
to the bus utilization definition (will be discussed later).
(10) A self-clocking, baseband signal modulation is to
be utilized.
(11) Word and message synchronization by unique wave-forms in the serial bit stream is to be used.
(12) The 16-bit information field in data words is to
be utilized.
(13) The message protocol and possible system applica-
tions should be decoupled and mutually independent.
(14) Hierarchical message addressing to the software
defined names will support the hierarchical system topology
and enhance the features of distributed bus control.
(15) The system is to allow both periodic and aperiodic
messages transfer, with aperiodic message being prioritized.
(16) Variable-length block and string messages transfer
should be possible.
(17) In addition to standard techniques to decrease
software complexity and costs (top-down design, partitioning
and modularity, higher-order-language usage, etc.), the sys-
tem software interface must match the hierarchical, distri-
"-". 145
SI
~ K: buted nature of the architecture that allows higher level of
"smart" hardware support.
(18) Some sort of dynamic reconfiguration is to be pro-
.5. vided to enhance system integrity so that recovery proce-
dures require little or no emergency reconfiguration in the
case of bus or terminal failure.
Error detection should be possible by parity per
word and by cyclic checksums at the end of longer messages.
Redundancy should be optional, rather than mandatory.
(19) Implicit broadcasting allowing transmissions from
one source to unknown number of receivers should be utilized
as the primary way of communication. At the same time,
- acknowledgments must be allowed even when multiple receivers
are involved, because the acknowledgments are essential for
the system integrity.
Comparison with the AE-9B "Future" Bus. Probably, this
is the right place to compare the "optimal" characteristics
listed above with the basic requirements, given in Table II
on the next page, for the "future high-speed data bus" as
recently defined by the AE-9B HSDB Subcommittee (Ref 5:54).
It is noticeable that they match in most of their key
issues. The variations are due to nuances in the definitions
of the essential characteristics, but there are not signifi-
cant conflicts between the two sets of characteristics. This
makes the "optimal" architecture, as defined in this report,
5$., 146
• ".-$. at least a candidate to be implemented in a hot bench and
further refined.
TABLE II
Prime Characteristics Required forFuture High-Speed Data Bus
Characteristics Goal
Distributed Bus Control To achieve maximum faultisolation and ease of system
_change
Data Rate Up to 20 Mbps throughput(dependent on technology
___available)
Baseband System Possible application tobroadband later
Linear Bus Multidrop, like MIL-STD-1553Coaxial or Fiber Optic Desire system to be independent
of physical bus medium charac-teristics
Protocols Contention, Token Passing (i.e.Polled/Round Robin), etc.
Integrated Data and Audio Standard intercom (ICS) andinteractive audio functionscombined on the bus. Bus to
*handle minimum of 20 stationsMessage (Protocol) Format Bus to handle all types of
message transfer and response(broadcast, request, andrequest with acknowledgment)
Message Length 4K words, maximum1 word, minimum
Word Length 16 bitsRedundancy Optional (single bus allowable)Bus Length Up to 150 metersBus Size Up to 64 terminalsError Detection Short Message - parity per word
up to and including 31 words;Long Message - 16-bit CRC formessages of 32 to 4K words
(Modified from Ref 5:54)
147
*0.*z"
~ The word "optimal" is put into quotation marks through-
out this report to point up the fact that what is optimal
for some particular applications is not necessarily optimal
for all others. In addition, the listed characteristics
represent, in a way, an envelope of the best features, that
are either taken from the analysis of very different and
sometimes mutually exclusive techniques or are just results
of some theoretical considerations. The consequence is that
all of the so defined characteristics cannot be easily
-. integrated in a single sys%'-em and some trade-off s are una-
voidable. In this way, the word "optimal" looses its primary
sense.
* In addition, any study on the subject would not be
complete without an attempt to synthetize a system based on
.1 the conclusions drawn from the study's results, regardless
of how difficult integration of the results could be. Also,
low level implementation details do affect overall system
characteristics to a greater extent than is usually recog-
nized.
These are all the reasons that an architecture, named
as the Self-Managing Multiplex System, is conceptually deve-
loped in this chapter. The description of the development
should also serve as a guide for the multiplex system "syn-
thesis issues" according to the goals defined at the begin-
ning of Chapter III (pp. 14-15).
~ 148
* 4 ' 4 . * ~ 1 .. -.: -c -- -C.L %
70 .2 -27
Self-Managing Multiplex System
Probably, the following preliminary design of the Self-
Managing Multiplex System (SMS) contains most of the "opti-
mal" architecture characteristics. Although some new ideas
are included, the concept avoids the pitfalls that are often
made when designing complex systems (Ref 76:658): (1) being
nothing revolutionarily new in the field, the SMS integrates
a variety of already implemented solutions, thus making
itself an evolutionary development from a successful current
* system design base. At the same time, the integration of
several existing techniques in a new context, based on the
analysis performed, has more powerful application possibili-
- ties; (2) the concept is not based either on "if a little is
good, a lot is better" logic, or on concentration of respon-
sibility only at one point in the design organization. It is
rather directed toward covering as much of desired characte-
ristics as the "ia" architecture is to have. The really
-' new ideas are introduced only when was necessary to resolve
conflicting requirements within the set of desired characte-
ristics.
Basic System Requirements. Any design of a data trans-
fer system must start from recognizing and defining the
mission requirements to be satisfied by the avionics archi-
tecture. For the purpose of this development, a hypothetical
aircraft is considered because of the lack of available data
149
for modern military aircraft.
According to the published data for the A-7D, carrier
based attack aircraft (Ref 49:89, 224-261), with its current
federated architecture similar to DAIS, the worst case total
data transfer loading is calculated as 84,351 bits/second to
be exchanged between 31 subsystems. This gives approximately
21 messages of ten 16-bit words with an average update rate-_
of 20 to 25 times per second.
The hypothetical aircraft is assumed to be capable of
handling at least 10 times more total data transfer loading
(843,510 bits/second), with 10 times more global variables
(messages) and double the number of connected subsystems
(i.e. , 210 and 64, respectively). The average rate of message
update is also to be doubled (about 50 times per second).
In order to exploit already discussed benefits, the
system is to be designed as a hierarchical, fully decentral-
ized architecture based on the "transparent contention"
control scheme for a time-division serial multiplex bus.
Transmission Medium. The system is to be independent of
physical bus medium characteristics. The relatively low data
transfer loading led to the decision to use twisted-shielded
pair as the transmission medium in the discussions that fol-
low. This type of transmission medium is a well-understood,
mature technology, that is inexpensive and reliable. In
.4" addition, some further benefits from a new concept can be
150
- * * .... ~ * % %*dr * '. . .*~:~'-~'
gained when such medium is used, as will be explained later.
However, the reader should be aware that the physical nature
of the particular medium chosen causes different effects
when this new concept is employed, and this will be pointed
out whenever applicable.
* Signal Modulation. The Manchester-II (Figure 11, p.30)
seems to be the most appropriate choice because of its wide
usage in the field. It is a standardized technique, employed
in almost every aircraft data multiplexing system today.
* There are quite a few manufacturers that are able to cope
with any hardware solution required when the Manchester - II
technique is used.
These facts will allow an inexpensive, large-scale pro-
duction of the hardware components for the SMS-based systems
in the future, protecting the investments into development
and existing production lines that already have been made.
Bus Allocation Scheme. The first of several new solu-
tions for the SMS concept is introduced in this subsection.
Namely, if the Manchester-II modulation is to be implemented
in a "transparent contention" bus allocation scheme, then
some solution should be found for the bus interface units to
compare their own current transmissions with actual signals
.1 on the bus and decide on whether to continue transmitting or
not.
Figure 51 depicts the transparent contention technique
151
implemented in a new way. An idealized bipolar Manchester
coding is depicted. Note that the technique will work equal-
a.' ly wt.ll if the unipolar Manchester is used, meaning that the
concept can be implemented with optical links as well (the
optical link technology, namely, requires some sort of uni-
polar coding because of the unipolar nature of optical
links: the presence of light or its absence are the only two
possible states on the bus).
In order to introduce the concept, assume that all ter-
minals connected on a single bus have the ability to trans-
mit and to receive at the same time. Assume, also, a distin-
guishable bus allocation signal is somehow generated and is
present on the bus. Further assume that each terminal,
QZ having some data to be sent starts transmitting at the bus
allocation signal and compares on a bit-by-bit basis what it
is trying to put on the bus with what is actually there. In
such a case, when the Manchester coded signals are generated
by separate sources, the bus will be a summation point. As a
* .. consequence, the majority of the terminals sending the same
bit will not notice any disagreement, since their receivers
will sense the bus voltage transition in the desired direc-
* tion. The terminals sending a different bit will notice the
disagreement and will stop transmitting, waiting for the bus
to become available again.
152
w1 0
15 1- I') z a)
-- - - - - - - - -- C
0Q)
* -o
imi
0 0a
cm0. 0H-
0 o
- _ I153
* * * . ... ----.-- -. 7 7%
.~ Such disagreements occur in Figure 51 at bit #3 and bit
#5 frames, thus terminals 4 and 3 stopped transmitting, re-
spectively. The figure shows that in this technique majority
rules and there is no inherent priority of one logical state
over another. This fact will be very useful when used in the
new concept of the self-correcting broadcast/acknowledgments
mode as discussed later.
An undefined status occurs in the case when equal
number of terminals try to put different bits on the bus at
the same time: the net result is zero voltage on the bus,
and by default, terminals sending "ones" will win over
those sending "zeros". The approach would work equally well
.ALL if "zeros" won in such a case, but design of the bus inter-
face unit (BIU) is simpler, as shown later, if "one" is
given such priority. At this point, there are two possible
solutions: either "ones" will be repeated by the sender(s),
as shown in Figure 51, or the receivers will automatically
replace this "equal contention dead bit" by a "one" (see the
modulation correction unit - page 110). In either case the
terminals sending "zeros" will stop their transmission.
In the proposed SMS design, the first solution is
preferred in spite of the fact that repetition of the bit
wastes one bit time. Namely, several successive "equal con-
tentions" may occur and, with the second solution, this
would make the BIU design more complicated (precise internal
~ ~:*154
.. 4 .. ~ clock circuits, prevention of automatic restoration when bus
is really silent, etc.). Thus, the wasting of one bit time
seems to be well-justified trade-off.
Considering the case depicted in Figure 51, terminal #1
wi.1l stop transmitting at the bit #10 frame, but terminal #2
will repeat its "one" and complete the message. The message
will be received by all receivers attached to the bus, in-
cluding the receiver of terminal #2. Note that a terminal
receives its own message over the bus and then stores it
-~ into its state information memory (SIM - see pages 127-129).
In such a way, all terminals are "equal" and this allows
each terminal continuous self checkof its own receiver. This
is very important and will be discussed in detail later on.
The number of terminals in competition is not a limi-
ting factor for the concept itself. Only one terminal at a
time completes its message without significant message over-
head required and without a bus controller of any kind.
There is no possibility that two terminals will ever try
to send identical words at the same time, as will be shown
in the next subsection describing the SMS protocol.
9a. Of course, the idealized Manchester waveforms as depic-
ted in Figure 51 are not feasible. The effects of line pro-
pagation dalays and attenuations may be relevant. Different
signal rise and fall time, as well as deviations in the zero
crossing point that vary from transmitter to transmitter,
*a~' ~155
- - ~ ~ ~ ~ ~ 7 '.-P S 70 j6 - J .F -.- .- .- ~- - .- .
* ***must also be taken into consideration. The noise environment
of an aircraft introduces additional problems.
Obviously, the on-line summation of such signals will
result in rather complex waveforms from which a receiver
must extract the bit information. This will be discussed in
.1 more details, once the concept is completely introduced.
i Message Protocol. The system message protocol is de-
fined in a data-variable addressed fashion. Namely, a mes-
.4 sage consists of two (or more) 16-bit words plus horizontal
parity bit for each word (Figure 52). The very first word
in each message contains a 12-bit field for the "names" of
variables to be exchanged in the system (i.e., altitude,
velocity components, etc.).
0 Eventual competition for the bus will be resolved based
on these names, which are unique for each variable in the
system. Thus, there is no possibility that two or more
terminals will ever have the right to send data at the same
time.
The length of the data words is determined based on the
discussion in Chapter III (page 43). In order to have uni-
fied word length in the system, the same 16-bit (plus pari-
ty) length is also chosen for the first, "contention resol-
ving and variables naming" word.
.4 Word synchronization is employed, rather than synchro-
nization only at the beginning of the message. This allows
.4., 156
better internal synchronization for all the BIUs and is
particularly important due to the reliability requirements
for the systems employing bit synchronization by extracting
a clock from the data stream. This is the case with the SMS
technique, where possible complex waveforms on the bus fur-
ther emphasize such a requirement.
1 2 3 4 5 6 7 8 9 10111213141516 17 20
",--'SYNC)1- " 1 1 1 1 1_ 1 1-'
A P
BY THE VARIABLE ADDRES IN GIVEN SYSTEM) INWS
" __ _ _ _ _ _ _ _ _ _ ____
• -"LAST DATA WORD (OR VERTICAL PARITY; EOT(4096 G(ALLOC.
• " " SIGNAL.)
Figure 52. The SMS Message Protocol
:5
."
: '. 15 7SYNC
%.ro0 lo
* ~. - ~ 7 a . K V- - 9 * . *; N _'V , 'r w . - U . -
:". .. The sync patterns are adopted from the MIL-STD-1553
approach, using the data sync (Figure 21,p.74) as inter-word
sync (IWS) and the command sync (Figure 20,p.74) at the end
of transmission (EOT) as the bus allocation time signal. All
data words within a message are separated by the IWS's.
Each terminal has a time-out circuit with a unique time
constant, triggering the EOT signal generator within the BIU
so that a silent bus will always have an EOT from time to
time, allowing the competition mechanism to start. Namely,
if there is no terminal wishing to transmit after the normal
message's EOT, the bus would be "dead" forever. This is re-
solved in that the terminal with the shortest time constant
will time out first and send a new EOT signal. If it fails
to do that, the next terminal will time out and so on. The
EOT is sent periodically as long as a new message is not
being transmitted. Each EOT signal, sent either as the nor-
mal ending of a message or as a separate signal, resets the
time-out circuits in all BIUs on the bus. The same mechanism
is also used at every initialization when the system is
powered up. Time constants are programmable, allowing the
system designer to adjust bus allocation signal periods
depending on the system requirements.
Refer back to Figure 52. A message containing data for
a variable always starts with a "zero" bit in the first word.
The second bit is -an acknowledgment bit ("one" means that
1 " 158
-*- .'" ~
the acknowledgment is requested and will be used depending
4 on the variable importance. Zero means no acknowledgment is
necessary). The 12-bit variable-name address field in bits
3 to 16 allows up to 4096 global variables to be used in the
system. The addressing method is intended to be in a hierar-
chical fashion (see page 46), allowing the system to be
partitioned and integrated into a hierarchical topology (see
page 17). The next two bits (15 and 16) are used in the
built-in-test (BITE) mode as the command signals, followed
in the last position (17) by a horizontal parity bit over
the first 16 bits. The first word, as already mentioned, is
used to resolve any possible competition and is ended by an
d IWS, followed by as many 16-bit data words as needed (usual-
ly one data word per variable is enough). All data words
within a message are separated by the IWS's. The last word
can be used as a vertical parity word allowing standard
self-correcting techniques to be implemented. Thereby, mes-
sages are of variable length, and this gives the system
designer greater flexibility. A completed message is always
ended by the bus allocation time signal (EOT), that replaces
the IWS.
The SMS concept assumes only broadcast mode for message
transfers, fully utilizing all the benefits of low overhead
in such a mode. At the same time, acknowledgments are used
in a new and very efficient way to fight transient errors as
.4. 159
-4
will be fully explained later on.
The broadcast mode, combined with the virtual memory
.technique (the SIM concept, pp. 127-129) offers a very effi-
cient way to handle real-time systems' requirements as poin-
ted out on the referred pages and will not be repeated here.
However, the state information memory technique is now
extended in the SMS concept by introducing so-called "book-
keeping words" (Figure 53) that are stored in the SIM along
with the variables.
1 2 151617 20
Ill Al P
VARIABLE ADDRESS IWS
" 8 12 1617 20
I I I I I I 1 1
Bu I I Bu 2 Bu # | 1 BU #I4 O
sages Figure 53. The SMS "Book-Keeping" Word
Namely, in order to reduce the bus load all of the mes-
sages will not be broadcast with the acknowledgment request,
which means that they are not of critical importance. The
acknowledgment mechanism, as will be seen, automatically en-
' " > 160
* , : ,',"4 ' '. , . . . -. .. . .* * , , , -' *' 4 ." ' "." - *';'' ", " , .'. " . '"-* • : 4* ".. . ." . ..
.~ ~ sures that all terminals (without exception) have received
good data. But, if such a possibility is excluded by the
system designer for a variable, this means that bad data for
such variable can be tolerated in the system if received
from time to time (for example, in some non-critical systems
such as in-flight monitoring equipment for post-flight ana-
4, lysis purpose). This further means that rejection of such
bad messages is not essential even when the error has been
recognized by the receiver, and that the sender does not
need to be notified immediately.
Having in mind more than one bus in the system due to
the reliability requirements, the purpose of "book-keeping
word(s)" is to allow different transient errors to occur
- - once on each of the buses, and the sender will be notified
only if the same error for the same variable occurred twice
on the same bus. In such case the "error signal message"
will be sent, competing for the bus in the usual manner. The
only difference between data messages and error signal mes-
sages is that error messages begin with a "one" in the bit
position 1, while content and the number of words describing
the errors on the buses are at the discretion of the desig-
ner. Figure 53 depicts the "book-keeping word" (error sig-
nal message) for a system designed for up to four buses.
There are four different errors that can be recorded for a
bus: parity, word length, data drop-out, and invalid IWS.
161
emVV
The use of the error signal messages is optional and
can be avoided if the system designer wants as small an SIM
as possible.
All messages can be acknowledged. This gives tremen-
dous reliability, but sacrifices data throughput, which
can be improved by more buses with greater cost, and so on.
The trade-off is the choice of the system designer.
It is also possible to design a system that repeats
critical data over several buses within the same time
frame. Repeating of the same critical message is a technique
employed in the Space Shuttle - the receiver's software
votes on such messages and uses them only if the software
determines that a good message has been received.
On the other hand, the use of error signal messages in
the SMS concept replaces a classical bus monitor with as
many active monitors as there are terminals on the bus. This
can be used as a suitable tool for mutual health-checking
among the terminals. It is true that errors are tied to the
variables and not explicitly to the terminals, but a varia-
ble is generated only by one terminal. Therefore, mutual
checking among terminals is involved. Sending of an error
signal message is the case when more than one terminal may
send the same message at the same time, but it has no impor-
tance since only the variable generating terminal will use
this message as a trigger for BITE self checking procedures.
~0~~ 162
All other terminals will reset book-keeping words for
the variable, suppress the same error message, and start the
process all over again.
Bus Interface Unit. This thesis work is intended to be
the definition of an "optimal" architecture. Any implementa-
tion is beyond the scope of this effort. However, as already
pointed out, low-level details usually do have implications
throughout the network architecture. This is the reason,
along with a need to illustrate the concept, that a proposal
for the four-bus BIU implementation is given in Figures 54
and 55. Note, please, that the proposal is given only as a
possible way to implement the technique and to show that an
actual design would also have to provide full hardware sup-
port to raise the system performance and lower the software
and maintenance costs.
The concept is illustrated by a BIU conceptual design
for four twisted-shielded buses, but the BIU can be connected
-, to optical links as well (the only difference would be the
bus coupler). The block diagram for the BIU is given in
Figure 54.
.4 The BIU is "smart", containing a processor with local
memory, a "state information memory" (SIM), a FIFO buffer
where messages are stored waiting to be transmitted, andI transmitter/receiver units (one or more for each bus) with
their own input and output buffers.
S 163
FIFO
CONTROLPROCESSING UNIT
SIMLOCAL MEMORY
0!
Figure 54. Bus Interface Unit Block Diagram
The fact that a BIU can transmit and receive simultan-
eously on different buses, along with the use of the trans-
parent contention concept and a special arrangement for
input and output buffers, allows the earlier mentioned fea-
tures of acknowledgments in broadcast mode and the tran-
sient error self-correcting capability, as will be explained
later.
Figure 55 illustrates one out of the four transmitter-
receiver units.
' 164
-'4Z
I
-4pJ
M.5-i
OI--O - o0
E-
- I a
t44
* A 165
The receiving channel consists of a limiter, a filter,
a Manchester decoder and validation circuit, and an input
buffer in a special serial-in, serial/parallel-out arrange-
ment.
The output channel contains a single-message parallel-in,
serial-out rotating buffer, a retransmission switch, a Man-
chester encoder and bit count generator, a shaping network,
and a bus coupler driver.
In normal process, the attached subsystem places mes-
sages into the output FIFO buffer which is common for all of
the four transmitters. The BIU's processor distributes the
messages from the FIFO buffer among the "empty" single-
message transmitters' output buffers and the messages are
ready to be transmitted. Each of the transmitters in a BIU
has a different message in its buffer and all can transmit
independently and simultaneously, but over the different
buses as connected. Thus, the discussion that follows is the
same for any of the four buses.
When an EQT occurs on the bus, the Manchester decoder
will trigger the internal clock generator. The transmission
control logic will at the same time eventually remove any
* existing abort signal and will apply the shift enable signal.
This will allow the content of the output buffer to be
clocked through the switch, encoder, shaping network, and
driver out onto the bus. Each bit sent out is also delayed
166
~. * ~.:before being applied to the comparator. This takes into
account the time needed for the sa-ne bit to arrive from the
other side of the comparator circuit. Any discrepancy bet-
ween comparator inputs will cause an abort signal for the
transmission control logic, and the encoder, as well as the
main switch will be blocked.
While clocking out the message, the output buffer will
at the same time rotate its content back into itself. When
an abort occurs, the transmission control logic will conti-
nue to allow rotation of the output buffer content until the
message is completely positioned for retransmission. Note
N that if there was no abort signal, this means that the mes-
sage won the competition for a bus and the message has been
completed. However, the same message is still in the output
buffer and can be retransmitted if needed.
If "equal contention" occurs, the Manchester decoder
and validation circuit in the receiving channel will gene-
rate a "one" for the Repeat AND gate and temporarily (for
one bit time) remove clocks, which will inhibit receiving
and transmitting. If the transmitter was trying to put a
"lone" on the bus when the "equal contention" occurred, the
repeat logic will cause this "one" to be retransmitted with-
out counting it as a new bit. At the same time, the "equal
contention" trigger and delayed "one" will override the
abort signal from the comparator. However, the encoder input
V- 167
will be blocked for one bit time, allowing retransmission.
If the transmitter was trying to put a "zr" on the bus
when "equal contention" occurred, the abort signal will cause
suppression of the transmission.
When the receiving buffer is full and if the message
was neither an error signal nor data with acknowledgment
request, the content of the buffer will be written into the
SIM, but only if no error has been discovered in that trans-
mission. At the same time, the four zero bits for error
signal will reset the variable's book-keeping word.
The variable will not be written into the SIM if an
error has been discovered; instead, the book-keeping word
and the 4-bit error signal for the message on the particular
0bus are compared. If the error discovered in that transmis-
sion is of the same kind as already written in the book-
Akeeping word, the CPU will flag the word and send it as the
error signal message when the BIU has nothing else to trans-
mit. This postponement of sending of error signal will give a
busy terminal a chance to not transmit the error condition
at all, because some of "free" terminals, if any, will do
that in the meantime and the bu sy terminal will simply reset
the word and remove the flag.
The book keeping word is always reset when a good
message is received. In that way, the subsystem using a
variable knows that the variable is properly updated and
.8 168
accurate if the book-keeping word for that variable is all
"zero". If one or more (different) errors are recorded, the
subsystem is warned that the variable has not been updated,
and will or will not use the variable as it is (depending on
- the subsystem and variable importance).
If the message was an error signal, writing into the
SIM will be inhibited. Also, the control BITE logic will
interrupt the CPU, and the CPU will decide either to per-
form BITE exercise if the variable was its own, or to reset
the book-keeping word.
Acknowledgments in Broadcast Mode. The possibility to
* fully acknowledge the reception of a message transmitted in
broadcast mode to unknown number of receivers is, probably,
0 the greatest advantage of the Self-Managing Multiplex Sys-
tem. To the author's knowledge, there is no similar solution
proposed so far. Usually, acknowledgments in a broadcast
mode are considered in the literature to be either unfeasi-
ble or to be unreliable. However, in the SMS concept such
* acknowledgments are not only possible, but they are also
used as a very efficient tool to fight transient errors.
Thus, system integrity is enhanced. The technique is as
follows:
Refer again to Figure 55. When a message with acknowle-
'C dgment request is received, writing into the SIM is inhibit-
ted, the CPU is interrupted, and the acknowledgment control
d ~' 169
logic is enabled so that this time the input buffer clocks
out its content through the main switch. Just before the re-
A transmission from the input buffer starts, each of the
acknowledgment control logic circuits (the retransmit con-
trol logic in Figure 55) will replace the acknowledgment bit
in the buffer by a "1zero"1 , so that only one retransmission
can occur at a time. Simultaneous retransmission by all ter-
minals at the same time and on the same bus means that mas-
sive contention will occur. The mechanism of receiving, com-
parison, and eventually stopping of transmission is as usual
and the net result is that receiver(s) which initially
received bad message during the original transmission will
* have the good message when the retransmission is completed.
Namely, in most cases transient errors will occur at the
minority of receivers, so that the majority will win during
K' the retransmission contention. The minority will stop their
transmissions, while receiving the good message this time.
This will correct any error(s) in the minority receivers'
buffers.
Of course, there is a possibility that exactly half or
even a majority of terminals will originally receive bad
~messages. In the acknowledgment retransmission process they
will win, but note that the sender is also receiving the
same retransmitted message. It knows what the good messagep is and in such case will simply repeat the message all over
170
%I
~ again. This is possible because (1) the good message is
still in its output buffer; (2) comparison on a bit-by-bit
basis of the retransmitted and the good message has been al-
ready made by the sender during the retransmission process;
and (3) the sender is only one which is responsible to send
either an EOT if the good message has been retransmitted or
an IW3 if repetition is needed. This repetition can be done
as many times as the system designer wishes - it depends on
how the self-fail time-out monitor is set. The sender is,
also, the only one which participates in the retransmission
by sending the message from its output buffer. This obvious-
ly increases the number of good messages on the bus, but
also allows the comparison of the good message with the
retransmitted one in the usual manner through the sender's
comparator. If an abort occurs for the sender's transmission
logic, the good message will continue rotating back into the
output buffer. Thus, when other terminals have finished the
retransmission of an obviously bad message, the sender will
have the good message ready and will send an IWS followed by
the message.
Therefore, there is a very small possibility that tran-
sient errors affect the system performance, because the var-
iables are written into the SIM only when it is almost ab-
solutely sure that they are good in all of the receiving
buffers. It is not likely that transient errors may occur
171
%4
*p\- ~ during the retransmission process, due to the fact that all
of the terminals are transmitting at the same time and the
signals on the bus are at or close to their maximal values.
One can argue that the technique has the same effect as
if the sender in a classical system simply sends the same
message twice or more times in a row. This is not true, be-
cause the sender without an acknowledgment never knows what
really has been received at the other end of the bus regard-
less of how many times the message is repeated.
In addition, in most cases the SMS acknowledgment tech-
nique will at the first retransmission correct misinterpre-
ted messages, if any. This is an obvious advantage over the
standard correcting techniques that utilize CRC word(s) sent
at the end of the message. Namely, the use of horizontal and
vertical parity bits ovpr the single words and the whole
message, respectively, is senseless in general if detecting
and correcting of more than three incorrect bits is required.
In contrast, with the SMS concept as many bits as the whole
message can be originally received wrong, but all the errors
will be automatically corrected during the acknowledgement
process.
There is no other technique which is capable to provide
such a feature in more efficient way.
On the other hand, the inherent priority that the
presence of light has over its absence, makes the SMS con-
172
*.. .*-~.cept, employed with optical links, more efficient in the
normal operation, but less efficient in the self-correcting
acknowledgment mode, when compared to the same system with a
twisted-shielded pair bus. Namely, in the normal operation
an "equal contention" will never occur on an optical bus. The
bits that are "one" (presence of light) always win over the
"zeros", regardless of how many terminals are trying to put
this "zero" (absence of light) on the bus. This fact allows
variables to be named in a prioritized order, which may be
very important for some systems (trigger messages could be
transmitted as soon as generated). However, since the majo-
rity does not rule on an optical bus, the self-correcting
feature of the broadcast acknowledgment is useful only for
the bits that were originally sent as "ones". In such a
- case, even if all terminals received this bit as a "zero"
the retransmission of "one" only from the sender's output
buffer will correct all the receiving buffers. But, if a "0"
was the originally sent bit and only one terminal has inter-
preted it as a "one", this terminal will win during the re-
transmission and will cause the sender to repeat the entire
message. Although it is not likely that the absence of light
will be interpreted as its presence, it is still possible
and the message may be required to be repeated. However,
since the bus bandwidth is not a critical factor when opti-
cal links are employed, the more often repetition is a well
~ 173
9,S
>* justified trade-off for enhanced reliability through the
sender's assuredness that all receivers have received the
accurate message.
System Network Topology. Figure 56 depicts a possible
network topology of the hypothetical avionics system. The
configuration is highly hierarchical in nature, decoupling
the system functions. This possibility for functional parti-
tioning is another feature of the SMS technique, allow-
ing a system design without the possibility that errors in
non-critical subsystems propagate and affect the flight-
critical functions. Note that there is no limitation for
growth of the system. Addition or deletion of subsystems, as
well as software changing or updating, does not affect other
subsystems. The same architectural concept can also be used
down to the lowest level when there are no subsystems attac-
hed to the BIUs - the processors in the BIUs are performing
the computational and other tasks.
A BIU is built by plugable, separated units. Such units
can be used anywhere in a system, even at the lowest (sensor
subsystem) level. This means a massive, cheap production.
The main goal of any standardization would be achieved in
that way, avoiding the trap of technological dependent stan-
dardization which can freeze possibilities for improvements
as technology advances.
174
4%t
-,
°L
bD
0
Cc 0S04
0
a)
1 :iE-4
-ao
IN175
-"" The interface for inter-bus communication is also built
by the same basic units (Figure 57) and the processor in
this case may be programmed to pass in either direction only
the variables needed by subsystems attached to upper and/or
lower levels. In such a case only FIFO buffers which are big
enough to accommodate all incoming variables would be pro-
vided, without a need for the SIM.
.I
'I _____ _T
XIt/RCv 91 XmIt/Rov 92 XmIt/Rcv 03 Xt/RovFIFO
FF CONTROL
SI M PROCESSING UNIT
LOCAL MEMORY
FIFO
XM- I t/C i A I tRv0 mItRv# mIt v0
Figure 57. Inter-Bus Interface Unit
Redundancy in the System. The number of buses at any
level is the designer's choice, as well as the number of
redundant BIU's transmitter/receiver units attached to the
176
*f , .. . . .. .. .. .. . . ....
7 -7 -7 Y% -... .
-- same bus(es). This redundancy for the BIUs is important when
subsystems of greater complexity are involved. Namely, while
the CRMmFCS technique of "task volunteering" (pp. 132-133)
provides at the lowest, subsystem level that there is no way
for faulty BIU to participate in message receiving, at the
higher levels this technique is not suitable because complex
subsystems cannot be replicated as many times as the task
volunteering technique requires. This fact hides a possible
-44' dangerous consequence: a faulty transmitter/receiver unit in
a EIU can cause that a message over a bus is missed. It is
true that distribution of the messages over different buses
is a pseudo-random process, such that the missed message
will be refreshed very probably at next sampling time by the
0 transmission over another bus where the BIU has a healthy
transmitter/receiver unit. In that way, if a missed message
has caused error propagation due to miscalculation based on
the old value of the missed variable, in the next time frame
it will be calculated correctly and the correction will
propagate through the same path, recovering the system. How-
ever, having the assumption of cheap BIUs, it is better to
have as many redundant transmitter/receiver units receiving
messages from the same buses as is affordable. The units can
then use the "volunteering" technique to compete for parti-
cipation in transmitting/receiving within given time frames.
- The faulty units will be simply removed from the system by
4 .y.."--177
giving up from the "volunteering". This also gives the
possibility that the BIUs processor (or other units by vo-
ting technique) can shut down the faulty unit if it fails to
do that by itself. (Continuous self-checking is already ex-
plained: a unit always receives its own message over the bus
and can monitor itself).
Another technique at the designer's disposal is to re-
configure the system if all of the available receiving units
for a bus are faulty. The BIU's processor could keep two ad-
ditional tables in its local memory. One table contains all
of the variables the terminal is reponsible to send out as
generated by the attached subsystem, and the second one will
contain d list of variables upon which the subsystem depends.
In the case that all transmitter/receiver units for a bus
are faulty, the terminal will simply send over other buses
the list of variables it depends on. In that way, the other
terminals will be informed about variables that should not
be sent any more over the particular bus. The terminals
generating these variables will append an inhibit flag for
the bus to the variables in their own generating tables and
will send these variables over other Lus(es). This allows
that the bus can still be used for other variables, without
affecting the system performance.
The number of buses used in a system is determined
based on the system requirements. A general guide should be
178
that the system will be safe if the number of faulty buses
is less than or equal to one-half of the initially available
buses. In the case of the hypothetical avionic system requi-
rements as defined at page 149, this translates into the
four buses required for the system. Namely, if 210 different
variables are to be exchanged with an average update rate of
50 times per second, the data transfer loading is
210 x 50 x 40 = 420,000 bi-s/second
where 40 bits are required to transfer a message containing
a single data word (page 157). Assuming the worst case, when
two-thirds of the variables are required to be acknowledged,
the load becomes 700,000 bits/second, which increases to
840,000 bits/second if the error rate is so high that every
0 other acknowledgment fails and the messages must be repeated
with further acknowledgments.
* If the rest of one-third of the variables are to be
covered by the book-keeping error signals that are transmit-
ted for each variable five times per second, then the load
increases by additional 70 x 40 x 5 = 14,000 bits/second and
the total load becomes 854,000 bits/second. The difference
to 1 Mbps of 146,000 bits/second is probably enough for the
purposes of trigger messages transfer, system reconfigurati-
on, the losses due to "equal contention", etc.
Assuming subsequent system growth up to twice the load
initially anticipated, the final result will be 2 Mbps of
179
-R 138 151 A STUDY OF OPTIMRL COMPUTER NETWORK ARCHITECTURE FOR 3/3DIGITAL AVIONICS SSEM .U. AIR FORCE INST OF TECH
WRIGHT-PAlTTERSON AFBO H SIHO OF ENG. M F KRLI
UNCLASSIFIED DEC 83 AFIT/GE/EEf'-D-6FG112 NmommmmmmmibiiEhhEEEhEE~hhHEWmmmllo
1.,
1 ',
• . • .1.
U4.
- LL
- wa-,1 1.5N=41.1M m 0
%- %
MINCP RESLUIO TETCHR
-~ A.TO#A BUEUO'rOIAD 6
Ah%
gross load in the system. That means that two always opera-
tional buses are needed in the system, which results in the
total of four buses, as has already been said.
Of course, the above calculation is very rough and ba-
sed on assumptions that may not be true for an actual sys-
e temn. Note, however, that the system is insensitive to loss
of one bus, completely safe if two buses fail, and still
usable if even three out of four available buses are not
operational (provided that message acknowledgments are re-
duced).
Bus Length. The signal propagation delays limit the
maximal allowable bus length. The same is true for any con-
cept, but in the SUS technique further complications are in-
volved due to the signal summation that results in quite
complex waveforms on the bus. There is a large number of
different waveforms that can occur as a consequence of
different signal propagation delays and attenuations, allow-
able zero crossing deviations, and different rise and fall
times for the generated signals. The situation is further
aggravated by the changing number of competing terminals.
A complete investigation of this phenomena requires
further research encompassing the total spectrum of possible
waveforms. An additional thesis effort, implementing the
Self-Managing Multiplex System via computer simulation tech-
niques is highly recommended.
180
~ ...~*.However, a limited number of cases that are investiga-
ted within this work led to the conclusion that there are no
problems at all when the concept is implemented with shorter
buses, such as the F-16 multiplex bus. If a longer twisted-
shielded bus is needed, then the existing technology for the
Manchester coding, decoding, and validation must be improved.
As a first, rough estimate for the generated signals, the
allowable zero crossing deviation should be limited to ±lOns
as well as the rise and fall times to the maximum of 100.0
ns. Also, the receivers must be capable to operate with the
incoming signals with a maximum zero crossing deviation of
i200.0 ns. With such characteristics, the allowable bus
length would be up to 150 feet, which is more than 8.5 times
longer than the F-16's bus, for example.
The maximum of 150 feet fcL a twisted-shielded bus is a
* conservative estimation, because none of the extenuating
circumstances is considered. Namely, the filtering effect of
just a few feet of cable significantly reduces the problems
* caused by the unfiltered waveforms (Ref 41:108). The twisted
pair shielded cable is essentially a distributed low-pass
filter. In addition, the limiter and the filter within each
'C receiver (Figure 55, page 165) are intended to shape the
incoming signals, so that the information can be extracted
from the direction of voltage transition. This transition is
only what is important for the Manchester decoding. Also,
CI 181
even current hardware is capable of handling signals varying
in a wide range from square to sine waves (Ref 65:27).
Note that the waveform deviation problems almost do not
exist in the acknowledgment mode. The simultaneous transmis-
sion of all terminals makes the signal distortion occur at
higher voltage levels, which will be cut of f by the recei-
vers' limiters. The greater the majority (which is normal in
the acknowledgment mode), the smaller the resulting signal
4 distortion. This fact adds more to the acknowledgment mode
importance.
Optical buses do not suffer from such problems, at
least not to the same extent. The bus length is limited only
by the requirement that the light presence within one bit
time frame must be continuous and must not extend into the
.1*~next bit time frame (to overlap with a new bit). Using the
convenient rule of thumb that the velocity of light is one
4., foot per nanosecond, it is easy to see that above require-
ments translate into a maximum bus length of up to 500 feet.
Number of Terminals. The electrical characteristics of
terminals that are feasible with today's technology also
expose some limitations on the number of terminals connected
if to a twisted-shielded pair. In order to calculate the allo-
wable number of terminals per bus for the hypothetical system
considered, a couple assumptions must be made first:
182
- ~ ~ ~ ~ ~ ~ ~ ~ - -7-A 7 - V FA. -. S.. * * 3. - *A' 9-
S".5..,
(1) The receivers' dynamic range should allow all ter-
minals to respond to Manchester-like signals with peak-to-
peak amplitude in the range of 0.86 volts to 14.0 volts for
transformer coupled stubs. The range is chosen as for ordi-
nary MIL-STD-1553 terminals (Ref 65:27), so that as much as
possible existing technology can be implemented.
(2) The line power loss is up to 1.5 dB/100 ft, as de-
fined in the MIL-STD-1553 (Ref 65:22). If the maximum of 150
feet in the bus length is assumed, then the worst case total
power loss on the bus is 2.25 dB.
(3) Referring again to the published data for existing
)systems (Ref 41:47 and 5:239), let the transmitter output
impedance be equal to the characteristic impedance (70 ohms)
of the bus, and the receiver input impedance be 5Kohm resis-
tive component at 1MHz.
Capacitor coupling is used at the transmitter and the
transformer coupling is used at each line receiver.
As has already been discussed (pp.20-25), the transfor-
mer coupling will provide very good common-mode,noise rejec-
tion while maintaining the bus in a balanced condition. The
capacitive coupling at the transmitters is chosen for this
system in order to provide a low impedance coupling to the
bus, while having the necessary dc isolation. The open
collector transistors in the bus-coupler drivers (Figure 55,
p. 165) are switched off when not being in the transmitting
183
~. ~ mode, isolating the transmitters from the bus. In fact, such
capacitor coupling has been utilized in a current system for
which enough data was available and this is the reason that
it is proposed for this system, too. For an actual implemen-
tation, the transformer coupling for both transmitter and
receiver is still more preferred.
With the described configuration of transmitter and
receiver coupling, the experimental results published by the
TRW Systems (Ref 5:239, Fig. 3) showed that the Manchester
2 signals voltage level received at the other end of a 500
feet long twisted-shielded bus, with a load of 64 terminals
and total power loss of 7.5 dB, was reduced only by one-half
with respect to the level of transmitted signals. Having a
considerably shorter bus (150 feet) and the receivers' dyna-
mic range assumed for the system described in this chapter,
it seems reasonable to assume that all transmitters are to
provide 1.5 volts, peak-to-peak amplitude for output signals
(for a comparison, this is the half of the B-52's voltage
levels - Ref 44:88, Table 9). Such low amplitude of the
output signals will decrease the power dissipation at the
transmitter's bus coupler driver, while still being above
the lowest allowable signal level at the farthest receiver
site (150 feet), even in the worst case when only one termi-
nal is transmitting.
Thus, if 64 terminals are assumed, the maximal voltage
184
AZ.,. .- -- .. - , * -... *..
AA," on the bus will be ±48 volts in the worst case when all
transmitters are transmitting the same bit at the same time.
This is less than 7 times as high as the upper limit of the
receiver's dynamic range. Today's technology allows the
receivers' active limiters to achieve such voltage attenua-
tion with relative ease (i.e., without extensive cooling
requirements).
Block and Trigger Message Transfer. The illustrative
design of the transmitter/receiver unit (Figure 55, p. 165)
only allows messages with a variable address and one data
word to be transmitted. Block-messages transfer transmis-
sion, if needed in an actual implementation, would require
a more complex design.
The optical links implementation of the SMS concept
allows a very efficient trigger message transfer. Thanks to
the inherent priority that the presence of light ("one") has
over its absence ("zero"), the proper addressing of trigger
messages will allow such messages to win any competition.
That means that a trigger message will be transmitted as
soon as it is generated. It is possible, even, to order
trigger messages themselves, although it is not likely that
two or more trigger messages will ever occur on the bus at
the same time.
With the twisted-shielded pair, ordering is not possib-
le, because the majority rules on such a bus. If the system
185
software complexity (to be discussed later) and the allowab-
le trigger message latency do not allow the worst case when
a trigger message waits until it is alone on the bus, then
hardwired connections for the trigger messages is recommen-
ded. If an actual system would require extensive hardwiring
for such connections, then the solution is to dedicate
P separate redundant multiplex buses for all trigger messages
in the system. The messages will compete for the bus in the
usual manner, but again, it is not likely that two or more
of such rare messages may occur at the same time.
Bus Efficiency. Calculated as defined for this report,
the bus efficiency for the SMS concept is constant in a nor-
mal operation and equals 40 percent. This is twice as much
as the MIL-STD-1553 efficiency, when compared with messages
containing one data word. The SMS system is also capable of
handling block-message transfers, although this is not em-
ployed in the conceptual design shown in this chapter. If
this was the case, the system would reach bus efficiency of
75% with just 16 data words in a message, while a MIL-STD-
1553 bus has such efficiency only as its maximum (32 data
* words within single message).
While a SMS optical bus retains the 40-percent constant
'S' efficiency in normal operation, the possibility of "equal
b.. contention" causes that a twisted-shielded bus may have this
efficiency in the range from 34.8 to 40 percent. The lower
186
* limit corresponds to the worst case of six such contentions
for 64 terminals trying to transmit at the same bus alloca-
tion signal.
Assuming, again, that two-thirds of the variables are
acknowledged with each other acknowledgment repeated and the
.4 book-keeping words for the rest of variables, then the SMS
bus efficiency drops down to 9.7 percent due to significant
overhead of doubled acknowledgments. Note, however, that the
assumption represents an extremely high error rate that is
not likely to occur in an actual system. The assumption is
made just because of eventual side effects with a twisted-
shielded bus, which should be thoroughly investigated within
o further research, as has been recommended.Bus Utilization and Message Latency. The bus utiliza-
tion and message latency for the SMS concept cannot be
calculated exactly, because the actual system specification
requirements are necessary for such calculations. Instead, a
.4'. discussion about another feature of the SMS concept follows:
The subsystem connected to its BIU provides data to be
transmitted only when the data has been changed since the
last time it was transmitted. In that way, the system be-
comes an information-only transfer system, not just a data-
transfer system. Namely, in conventional systems which are
* data-transfer systems, a bus can be just as busy sending or
receiving old data already known by all terminals as trans-
187
4.7
ferring important information (Ref 61:690). Information-only
transfer will increase considerable the data throughput.
Repetitive updates of static information consume much
of the available bus bandwidth. A more meaningful characte-
ristic to calculate for an information-only transfer system
would be the bus utilization defined as ratio of useful data
transferred versus the total number of bits used (Ref 90:43).
A comparison, based on this bus utilization ratio defini-
tion, between all of the systems described in this report
and the Self-Managing Multiplex System, would very probably
significantly reduce the performance figures of all other
systems since much of data they transfer is redundant.
Of course, the SMS bus time must still be allocated for
the worst case when all variables do change, but the bus
will become idle only when all of important information al-
u ready has been sent. This is far better than the case when
an important message waits for useless data to be sent
first.
Obviously, a significant decrease in information laten-
cy (not just message latency) is the primary advantage of
the information-only systems. Thus, some in-line delay inhe-
rent to SMS hierarchical topology due to the inter-bus
communication (Figure 56, page 175) is more than sufficiently
compensated by the benefits of information-only transferring.
188
..-
Software Interface. The software engineering aspects
are on the other side of coin and are as important as the
architectural and hardware support considerations.
One could argue that there is the possibility for a SMS
variable to be ready in the output buffer but wait forever,
constantly losing the competition. However, the worst case
calculation made on page 179 for the hypothetical system,
shows that each variable has its chance 50 times per second
on the average, which is exactly what was needed for the
system. In normal operation, this total load for the worst
case is less than 50 percent of the available 4 Mbps in a
SMS system with four twisted/shielded buses. Obviously,
oeach variable will sooner or later get its chance due to the
fact that the buses are never idle if there is some data to
be sent. The situation is further improved by the informa-
tion-only transfer capability.
The probelm is how to assure that the variables occur
on the bus when needed by the other subsystems (the SMS ap-
proach assumes only broadcast mode, and a source never asks
for information).
In a command/response system, as well as in polling
techniques, the capability for informatiqn request is paid
for by system inflexibility and by high message overhead.
A very useful approach to the problem is known as the
"quantized software" technique (the "software millimodules"
189
3 ~~ ~~~- . .A - * ~ W
in the CRMmFCS, p. 130-131). It gives an opportunity to know
almost exactly when a variable will be ready for transmis-
sion, but such an advantage is paid for by wasting some am-
ount of time when a processor idles waiting for the next
time milliframe.
%tion when a new variable arrives and uses this "event" as Thaeetdie"ssesreur rcso nerp
trigger for the task execution. Although very suitable for
real-time systems, this technique requires more complicated
hardware.
For the SMS concept it is assumed that the subsystems
are responsible to send only the data that has changed since
the last time sent. This means that the other subsystems
41 will use the value already stored in the SIMs. In fact, each
* subsystem is constantly executing its program as a recursive
loop by first performing all necessary readings out of the
SIM, then executing the program as a whole or partially as a
module. The results are then compared with the previous
values and will be put into BIU's output FIFO buffer only if
the variable has been changed since the last time updated.
The program execution then proceeds to the next module, Of
course, it may happen that a variable needed for a previous
software module arrives with a new value while the processor
is already executing subsequent module(s), but that is not
of critical importance unless the time expired from when the
190
~. variable has been generated to the time when it is used is
important (as in navigation systems, for example). This
problem, however, is already solved by time-tagging to such
data (Ref 49:220) and the same solution is also assumed to
be used in the SMS concept-based systems. What is important
for the system designer is to assure that the entire system
software is designed in a way giving no possibility that any
variable can be transmitted more than once before actually
used by each of the corresponding subsystems. If such a
possibility is unavoidable, the solution is in "quantized
software", especially at lower, subsystem levels where sac-
rificing the bus throughput (which usually is not critical
at these levels) for simpler software is well justified.
Anyhow, both possibilities do exist and are, again,
open to the designer's choice.
%7.
4 191
A* . * * * * 4.
VI. Conclusions and Recommendations
This report has been intended to define the "optimal"
computer network architecture for future digital avionic
systems. The author identified more than five hundred ref e-
rences that show how much attention the subject has received
in recent years. The survey of available literature as given
in Chapter II, is one of the results of this effort and can
- ~r. help further researchi in the field.
In order to define the characteristics of optimal ar-
chitecture, the current theoretical knowledge is documented
in Chapter III, where an attempt is made to sort out the
essential figures of merit of any avionics system architec-
ture. The key issues, as defined in that chapter, are used
in Chapter IV to evaluate fourteen different approaches to
the same problem of data multiplexing in avionic systems. In
that way, the current system design base is also documented
V within a single report.
4 Chapters III and IV represent the second result of the
study and can be eventually used in further research to
* establish some ordered and weighted criteria for evaluation
of any architecture that may be proposed in the years to
come. The only similar attempt, made in the reference 49, is
not elaborated to the extent which allows an automated eva-
luation (by a computer simulation technique, for ( xqmple).
192
~ Such research is recommended and this author hopes that the
Chapters III and IV of this thesis can serve as a starting
point should the recommendation be accepted, in spite of the
conclusion (page 15) that the uniquely defined criteria to
evaluate all possible solutions can hardly find the univer-
sal application.
I The characteristics of the optimal computer network
architecture for future digital avionic systems are defined
in Chapter V.
Briefly, the optimal architecture must reflect the re-
sults of the successful current system design base, invol-
ving advantages of proven techniques. The architectural
approach should be reliable, simple, and general enough to
-'S be standardized for use throughout the aerospace industry.
The standardization should be possible from the lowest avio-
* nics subsystem up to the global system level. The archi-
tectural philosophy should also leave the system designer
free and technologically independent to implement the con-
* cept according to the particular subsystem or overall avio-
nic system requirements, allowing at the same time a possi-
bility for subsequent growth and multi-mission applications
of the system with relative ease.
The author believes that the described Self-Managing
Multiplexing concept approaches an "optimal" computer net-
work architecture for avionics systems, as defined above.
193
The concept includes a variety of solutions already
implemented, but integrated in somewhat new ways, giving
F features that cannot be found in the other techniques. it
also offers great flexibility to system designers, without
technology dependence, and enhances the fault-tolerance to a
high level.
The normal procedure of dynamic reconfiguration requi-
res no emergency reconfiguration in the case of failures.
The technique is well-matched to distributed architectures
and the hierarchical topology of the system allows functio-
nal partitioning both in hardware and software.
Fast and dynamic, the transparent contention bus real-
location scheme allows the bus utilization in any sequence
and with any percentage of bandwidth. The information-only
transfer concept adds more to these features.
The benefits are not achieved without cost: high level
hardware support is needed. However, it could be considered
as an acceptable trade-off, especially if the concept finds
its place in future avionics standards and allows massive,
cheap production. This will also simplify logistics by com-
mon modules maintenance. In addition, as pointed out several
times throughout this report, more sophisticated hardware
p. combined with a high bus bandwidth in an advanced architec-
ture will raise the system performances and decrease soft-
ware complexity and costs at the same time.
194
% * ..
-_ ,I .. '-t. . -*** - . * q .. .q - . - i . - - °' - -" - m
.'- The Self-Managing Multiplex System is introduced only
conceptually and requires further engineering design, refi-
nements and implementation trade-offs on a hot-bench. A
laboratory testing of variety of different ideas integrated
in this concept is desirable, but a computer simulation of a
SMS-based system with twisted-shielded bus should be perfor-
med first to lower the risk of hardware investment in a new
technique. A complete investigation of the total spectrum of
possible bus waveforms is highly recommended and can be
performed within additional AFIT thesis work.
The intriguing features of the newly introduced acknow-
ledgment technique in a broadcast mode with automatic self-
correcting of burst transient errors, could be very useful
if proven in practice. As far as the author's knowledge goes,
there is no similar proposal made so far and there is no
other technique that offers so efficient a way to fight tran-
sient errors; therefore, feasibility testing is really
necessary. The technique can be tested even outside of a
SMS-based system, on the CRMmFCS hot bench that exists in
the AFWAL/WPAFB. Modifications to the bench are straightfor-
ward and, at least, this result of the study seems to deser-
ve some sort of testing in practice as soon as possible.
195
Bibliography
1. Adams, E.S. and B. Clausing. "Distributed AvionicsProcessing Using ADA," Proceedings of the NationalAerospace Electronics Conference, 979-983. Dayton,Ohio: Dayton Section IEEE. May 1983.
2. Allen, R. et al. System Avionic Architectures for RPVs.Technical Report, Texas Instruments, Inc., Dallas, Texas.
April 1977. (AD-A041502).
3. Air Force Systems Command - ENASD. The MIL-STD-1553Multiplex Applications Handbook, WPAFB, Dayton, Ohio,May 1980. (AD-A119529).
4. Barnett, P.R. "Implementation of Broadcast Messages andAcyclic Data Transfer Techniques in a Multibus BasedAvionic System," Proceedings of the National AerospaceElectronic Conference, 625-632. Dayton, Ohio: Dayton
Section IEEE. May 1983.
5. Bello, L.M. and T.C. Berg. "LSI Data Bus for Avionics,"Proceedings of the National Aerospace ElectronicsConference, 238-244. Dayton, Ohio: Dayton Section IEEE.May 1973.
6. Boose, E.F. "Lessons Learned Through a MIL-STD-1553Time Division Multiplex Bus," Proceedings of theNational Aerospace Electronics Conference, 634-641.Dayton, Ohio: Dayton Section IEEE. May 1975.
7. --- . "Shielded Twisted Pair as a Transmission Mediumin a Multisubscriber Time Division Multiplex System,"Proceedings of the National Aerospace ElectronicsConference, 648-655. Dayton, Ohio: Dayton Section IEEE.May 1975.
8. Booton, W.C. A Conceptual Definition Study for a DigitalAvionics Information System (Approach ),L Volume I.Technical Report, General Dynamics Convair AerospaceDivision, Fort Worth, Texas. October 1973. (AD-771736).
9. ------- A Conceptual Definition Study for a DigitalAvionics Information System (Appr. I),Vol. II. TechnicalReport, General Dynamics Convair Aerospace Division,Fort Worth, Texas. October 1973. (AD-771756).
196
UI~ 4 L~' ~.°.i~ T. ... . ; ' . - ,. . 1 . , .- 7..- ° -. -
10. Brodnax, C.I. A Conceptual Definition Study for aDigital Avionics Information System (Approach II),Volume I. Technical Report, Texas Instruments, Inc.,Dallas, Texas. March 1974. (AD-780581).
11. A Conceptual Definition Study for a DigitalAvionics Information System (Approach II), Volume II.Technical Report, Texas Instruments, Inc., Dallas,Texas. March 1974. (AD-780582).
12. ------- A Conceptual Definition Study for a DigitalAvionics Information System (Approach II), Volume III.Technical Report, Texas Instruments, Inc., Dallas,Texas. March 1974. (AD-780583).
13. Bussell, J.F. "A Concept of Digital Avionics,"Proceedings of the AGARD Conference "Inertial NavigationSystems and Components", 283-295. Braunschweig,W. Germany, May 1968.
14. Calhoun, M.D. and B. Khatirzad. A Study of Avionics TimeDivision Multiplex Bus Simulation. Technical Report,Mississippi State University, Mississippi State,Mississippi. December 1980. (AD-A097375).
15. Chien, R.T. and L.J. Peterson. Optimized ComputerSystems for Avionics Applications. Technical Report,Coordinated Science Laboratory, University of Illinois,Urbana, Illinois. February 1980. (AD-A089570).
16. Cook, M.J. et al. Digital Avionics Information System(DAIS): Development and Demonstration. Technical Report,TRW Defense and Space Systems Group, Redondo Beach,California. September 1981. (AD-A107906).
17. Cotter, G.E. et al. "Avionics Multiplexing with SmartTerminals," Proceedings of the National Aerospace
-Electronics Conference, 672-678. Dayton, Ohio: DaytonSection IEEE. May 1975.
18. Courter, J.G. "Application of Multiplexing to the B-iAircraft," Proceedings of the National AerospaceElectronics Conference, 351-354. Dayton, Ohio: DaytonSection IEEE. May 1973.
197
,-* ' *,**'* 9.V .* ********* *~,. %
19. Crossgrove, W.A. and J.D. Blair. "Categorization ofData Bus Architectures for Avionics and Flight ControlSystems," Proceedings of the National Aerospace Elec-
tronics Conference, 406-411. Dayton, Ohio: DaytonSection IEEE. May 1980.
20. -------. "Distributed Systems: The Next IntegrationMethod," AIAA Paper No. 77-1484. (1977).
21. "Development and Applications of MIL-STD-1553,"SAE Technical Paper Series, No 801142. October 1980.
22. Cunningham, John M. Investigation of Functional Common-ality of Avionics Systems in Naval Aircraft. Technical
* Report, Naval Air Development Center, Warminster,Pennsylvania. September 1981. (AD-A105503).
23. Czuchry, A.J. et al. MID-1980's Digital AvionicsInformation System Conceptual Design Configuration.Technical Report, Dynamics Research Corporation,Wilmington, Massachusetts. July 1976. (AD-A032137).
24. Davies, D.W. et al. Computer Networks and TheirProtocols. John Wiley & Sons, 1981.
25. Davis, Richard T. "Solutions Considered for Future HighSpeed Military Bus," Defense Electronics 5 (15): 48-62.May 1983.
26. Denery, D.G. et al. "Design, Development, and FlightTest of a Demonstration Advanced Avionics System," NASATechnical Memorandum 84323. January 1983.
27. Denny, M. and T. Planbeck. "Active Memory Subsystems:An Architectural Approach to Design Recoverability,"Proceedings of the National Aerospace ElectronicsConference, 1279-1282. Dayton, Ohio: Dayton SectionIEEE. May 1980.
28. Dunn, G.L. and P. Leong. "Conceptual Design of anIntegrated Power and Avionics Information System,"Proceedings of the National Aerospace ElectronicsConference, 954-962. Dayton, Ohio: Dayton SectionIEEE. May 1981.
29. "Real Time Monitoring and Analysis of aMIL-STD-1553A/B Multiplex Data Bus," Proceedings of theThird AIAA/IEEE Digital Avionic Systems Conference,
145-149. Dayton, Ohio. October 1979.
IJ " -;~ 198
% % A
30. Dworkin, B. "The Avionics Tree - A SystematizationS.of System Avionics," Proceedings of the National
Aerospace Electronics Conference, 89-95. Dayton, Ohio:Dayton Section IEEE. May 1976.
31. Elmore, Vicki L. et al. "Squeezing 1553 Bus Into theMSRAM," Defense Electronics, 104-113. August 1983.
32. Engel, H.E. et al. Digital Avionics Information System(DAIS): Current Maintenance Task Aialysis. TechnicalReporit, Dynamics Research Corporation, Wilmington,Massachusetts. October 1976. (AD-A035683).
33. Friedman, S.N. "MIL-STD-1553 Dynamic Bus Controller/Remote Terminal Hybrid Set," Proceedings of theNational Aerospace Electronic Conference, 639-644.
4. Dayton, Ohio: Dayton Section IEEE. May 1983.
34. Gaarder, T.N. The Theory of Digital Processing ofAStohastic Signals for Use in Avionic Systems on-Board
Air Force Aircraft. Scientific Report, University ofHawaii, Honolulu, Hawaii. March 1971. (AD-719822).
35. Gangl, Ervin C. "Avionic Systems Integration UsingDigital Computers," Proceedings of the AGARD Conference:"Automation in Manned Aerospace Systems." Dayton, Ohio.0 October 1972.
36. "Digital Systems Integration Techniques, the Keyto Affordable Avionics," AIAA Paper No. 77-1492.(1977).
37. Gifford, C.A. "A Military Standard for Multiplex DataBus," Proceedings of the National Aerospace ElectronicsConference, 85-88. Dayton, Ohio: Dayton Section IEEE.May 1974.
38. Gregory, J.G. "System Architecture for AircraftAvionic and Electrical Systems," Proceedings of theNational Aerospace Electronics Conference, 30-35.Dayton, Ohio: Dayton Section IEEE. May 1973.
39. Griffin, Edward L. "Standards and Integrated AvionicDigital System Architecture," Proceedings of the SecondAFCS Standardization Conference, 563-581. Air ForceSystems Command, WPAFB, Dayton, Ohio. December 1982.
40. Gross, J.P. "Data Bus Techniques for Digital Avionics,"Proceedings of the National Aerospace ElectronicsConference, 230-237. Dayton, Ohio: Dayton SectionIEEE. May 1973.
199V°
51. Kuo, F.F. Protocols and Techniques for Data Communica-*[ .. tion Networks. New Jersey: Prentice-Hall, Inc., 1981.
52. Lambert, J.W. and J.P. Green. A Comparative Study ofData Bus Systems. Technical Report, Intermetrics Inc.,Cambridge, Massachusetts. Marc 1978. (IR-270 undercontract N62269-77-M-5298 for NADC).
53. Larimer, Stanley J. "A Solution to Bus Contention in aSystem of Autonomous Microprocessors," Proceedings ofthe National Aerospace Electronics Conference, 309-317.Dayton, Ohio: Dayton Section, IEEE. May 1982.
a.
54. AFWAL/FIGC, WPAFB, Dayton, Ohio. Private con-versation. 18 July 1983 and 30 August 1983.
55. A Continuously Reconfiguring Multi-Microprocessor Flight Control System. Technical Report,
NFlight Dynamics Laboratory (AFWAL/FIGC), WPAFB, Dayton,Ohio. May 1981. (AD-A101412).
56. Leib, John E. "Application of 1553B to MSRAM - A SystemLook," Proceedings of the Second AFSC StandardizationConference, 295-304. Air Force Systems Command,Dayton, Ohio. December 1982.
57. Mahler, B.B. An Analytical Examination of Impedanceand Signal Power Levels along a Multiplex Data Bus.Technical Report, Mitre Corporation, Bedford, Massa-chusetts. March 1975. (AD-AO08498).
58. ------- A Comparison of Time Division Multiplex DataBuses Based on Computer Simulation. Technical Report,Mitre Corporation, Bedford, Massachusetts. May 1975.(AD-A012305).
59. Mason, R.C. et al. Digital Avionics Information System(DAIS): Development and Demonstration. TechnicalReport, TRW Defense and Space Systems Group, RedondoBeach, California. March 1979. (AD-A068438).
60. McCartney, R.I. and R.E. Phillips. "A Modular Approachto MIL-STD-1553 Simulation Support," Proceedings of theNational Aerospace Electronics Conference, 872-877.Dayton, Ohio: Dayton Section IEEE. May 1981.
61. McDonald, J.E. "An Architecture for Event-Driven Real-Time Distributed Computer Systems," Proceedings of theNational Aerospace Electronics Conference, 688-694.Dayton, Ohio: Dayton Section IEEE. May 1983.
*201
W 4.
41 ------- Techniques for Interfacing Multiplex Systems.'.1 .', Technical Report, SCI Systems, Inc., Huntsville,
Alabama. February 1981. (AD-A101457).
42. Haley, A.L. "MIL-STD-1553 Validation Test Results,"Proceedings of the National Aerospace ElectronicsConference, 747-749. Dayton, Ohio: Dayton SectionIEEE. May 1980.
43. Harris, R.L. and E.E. Jones. "Fault Tolerance Applica-tions to Future Military System Avionics," Proceedings ofthe National Aerospace Electronics Conference, 412-419.Dayton, Ohio: Dayton Section IEEE. May 1980.
44. Hawkins, R.D. "Development Considerations for Real-Time,* Aircraft/Avionic Distributed Computer Control Systems,"
Proceedings of the National Aerospace ElectronicsConference, 586-589. Dayton, Ohio: Dayton SectionIEEE. May 1983.
45.--------. "Buffered Receptor, Avionics IntegrationNetwork (BRAIN) -- A Concept Proposed for Memory ManagedAvionics," Proceedings of the National AerospaceElectronics Conference, 659-668. Dayton, Ohio: DaytonSection IEEE. May 1983.
0 46. Hecht, Herbert and Myron. Fault Tolerance, Reliabilityand Testability for Distributed Systems. First InterimTechnical Report, SoHar Inc., Los Angeles, California.February 1983. (RADC-TR-83-36).
47. Husbands, C.R. "Microprocessors in Ariborne TimeDivision Multiplex Terminals," Proceedings of theNational Aerospace Electronics Conference, 642-647.Dayton, Ohio: Dayton Section IEEE. May 1975.
-p 48. ILC Data Device Corporation. MIL-STD-1553 Designer'sGuide. Bohemia, New York, 1982.
49. Jensen, D.E. et al. The Impact of Wideband MultiplexConcepts on Microprocessor-Based Avionic SystemArchitectures. Technical Report, Honeywell Ssytems andResearch Center, Minneapolis, Minnesota. February 1978.(AD-A057878).
50. Klos, Larry C. "Protocol for Modular Avionics Archi-tecture Serial Multiplex," Reported at the National
-N Aerospace Electronics Conference. Dayton, Ohio:Dayton Section IEEE. May 1983.
200
.*, 62. McLaren, B.D. "B-I Avionics Data Multiplexing System, -..,, (AMUX)," Proceedings of the National Aerospace Electronics* "Conference, 205-212. Dayton, Ohio: Dayton Section
IEEE. May 1975.
63. McSharry, M.E. and L.A. Smith. "Contention InformationTrasnfer Systems Analysis," Proceedings of the NationalAerospace Electronics Conference, 402-426. Dayton,Ohio: Dayton Section IEEE. May 1980.
64. Meyer, D.H. and D.R. Parr. "A MIL-STD-1553 FlexibleInterface Device and Applications," Proceedings of theNational Aerospace Electronics Conference, 618-624.Dayton, Ohio: Dayton Section IEEE. May 1983.
65. MIL-STD-1553B: Aircraft Internal Time Division CommandResponse Multiplex Data Bus. 21 September 1978.
66. MIL-STD-1760: Aircraft/Store Electrical InterconnectionSystem. 1 July 1981.
67. Moran, John F. "A Bus Oriented Modular Avionic Archi-tecture," Proceedings of the National AerospaceElectronics Conference, 698-702. Dayton, Ohio: DaytonSection IEEE. May 1980.
68. Morgan, R.D. and R. Bellem. "Pave Pillar: A MaturationProcess for an Advanced Avionics Architecture,"Proceedings of the Second AFCS Standardization Conference,67T-690. Air Force Systems Command, Dayton, Ohio.December 1982.
69. Mrazek, J.G. "Flight Test and Demonstration of DigitalMultiplexing in a Fly-By-Wire Flight Control System,"
*Proceedings of the National Aerospace ElectronicsConference, 192-197. Dayton, Ohio: Dayton Section
-I IEEE. May 1973.
70. Mueeoq, R. and J.L. Perry. "Multiplexing in Aircraft-Subsystems," Proceedings of the National Aerospace
Electronics Conference, 343-450. Dayton, Ohio: DaytonSection IEEE. May 1973.
71. Newcomb, R.W. Theory of Functional Networks and Micro-systems for Avionics System. Scientific Report,Stanford University - Electronics Laboratory, Stanford,California. March 1970. (AD-705677).
72. Paz, Israel M. "Advances in Optimal Routing Through Com-puter Networks," NASA Technical Report R-475. September
S.%y 1977.
202
" 73. Pensworth, F.L. "Review of Different Approaches to theStandard Interface Problem," Proceedings of the NationalAerospace Electronics Conference, 245-252. Dayton,Ohio: Dayton Section IEEE. May 1973.
*,74. Porubcansky, Cynthia A. (Editor). Proceedings of theSecond AFSC Avionics Standardization Conference.HQ ASD/ENAS, WPAFB, Ohio. December 1982. (ASD-ENA-TR-82-5031).
75. Reese, I.R. and D.R. Porter. "An Airplane Fiber OpticMultiplex Data Bus Design and Feasibility DemonstrationSystem," AIAA Paper No. 77-1475. (1977).
76. Rosenkoetter, E.A. and R.C. Burns. "Advanced SystemIntegration Demonstrations (ASID) for the 1990's,"Proceedings of the National Aerospace ElectronicsConference, 651-658. Dayton, Ohio: Dayton Section IEEE.May 1983.
77. Ruggles, R. and E.M. Scott. "Developments in AircraftDigital Systems," Proceedings of the AGARD Conference:"Automation in Manned Aerospace Systems". Dayton,Ohio. October 1972.
78. Sanducci, Anthony F. "MIL-STD-1553 Bus Controller andMultiplex Remote Terminal," Proceedings of the NationalAerospace Electronics Conference, 1311-1318. Dayton,Ohio: Dayton Section IEEE. May 1981.
79. Scarpino, F.A. and J.G. Weber. "A DAIS ArchitectureUtilizing Microprocessors," Proceedings of the NationalAerospace Electronics Conference, 82-88, Dayton, Ohio:Dayton Section IEEE. May 1976.
80. Semcor Inc., Moorestown, New Jersey: A Comparison ofDistributed Control and Command/Response MultiplexSystems. Technical Report under contract N62269-74-C-0761. January 1975. (AD-A025105).
81. Smith, A.L. et al. Advanced Avionic Systems forMultimission Applications. Technical Report, BoeingMilitary Airplane Company, Seattle, Washington.October 1982. (AD-A121793).
82. Spencer, Larry T. "Fiber Optics for the Future - Wave-length Division Multiplexing," Proceedings of the SecondAFSC Avionics Standardization Conference, 871-888.H'SD/ENAS, WPAFB, Ohio. December 1982.
203
83. Sunshine, C.A. Communication Protocol Modeling.- . Artech House, Inc., 1981.
84. Tavora, C.J. and J.R. Glover. "The RLU - An AdvancedRemote Terminal for Avionic Systems," Proceedings of
N the National Aerospace Electronics Conference, 1271-1278. Dayton, Ohio: Dayton Section IEEE. May 1980.
85. Treacy, J.J. "FAA Certification Criteria for Criticaland Essential Digital Ssytems," SAE Technical PaperSeries, No. 811060, October 1981.
86. Verdi, James S. High Speed Multiplex Bus Protocol Study.Technical Report, Naval Air Development Center, Warminster,Pennsylvania. June 1980. (AD-A102113).
87. Weber J.G. et al. "DAIS: Design and Performance,"*AIAA Paper No. 771499. (1977).
88. Wiegand, R. "Digital Data Transform Applications,"Proceedings of the National Aerospace ElectronicsConference, 703-709. Dayton, Ohio: Dayton SectionIEEE. May 1980.
" 89. Wigle, G.B. Spare Memory and Timing Parameters inAvionics Computer System Requirements. MS Thesis,
-Air Force Institute of Technology, WPAFB, Dayton, Ohio.December 1977. (AD-A056521).
90. Wilson, Dayton H. "A Polled Contention Multiplex Sys-tem Using the MIL-STD-1553 Protocol," Proceedings of theFourth AIAA/IEEE Digital Avionic Systems Conference,264-270. Dayton, Ohio: Dayton Section, IEEE, June 1981.
91. Windham, W.A. "An Application of Simulation to theDesign Formulation of a Helicopter Integrated MultiplexSystem," Proceedings of the National AerospaceElectronics Conference, 682-687. Dayton, Ohio: DaytonSection IEEE. May 1983.
92. Wright, R.E. "Avionics System Architecture," Principlesof Avionics Computer Systems (Chapter §j, December 1974.
4 TGARD-AG-183).
204
S - -. .... ....... . .............. ......
VTA
Mensur F. Krilic was born on September 3, 1942, in
4, Sarajevo, Yugoslavia. He graduated from the First Classical
High School in Sarajevo in 1961 and from the Yugoslav Air
-~ Force Technical Academy in 1964, receiving a commission in
the Yugoslav Air Force. He also graduated from the Faculty
of Electrical Engineering, University of Belgrade, Yugoslavia,
in 1976, receiving a Bachelor of Science degree in Electrical
Engineering and has been elected as an assistant-professor
in the field of Aircraft Electrical and Electronic Equipment
in 1981 at the same Faculty.
He entered the United States Air Force Institute of
Technology in June 1982.
00
UNCLARS T FT RDmCURITV CLASSIFICATION OF THIS PAGE
IREPORT DOCUMENTATION PAGEP..., REPORT SECURITY CLASSIFICATION lb. RESTRICTIVE MARKINGS
SECURITYUNCLASSIFIED2iL SECURITY CLASSIFICATION AUTHORITY 3. DISTRIBUTION/AVAILABILITY OF REPORT
2b. DECLASSIPICATION/DOWNGRAOING SCHEDULE Approved for public release;distribution unlimited.
4. PERFORMING ORGANIZATION REPORT NUMSER(S) 5. MONITORING ORGANIZATION REPORT NUM8ER(S)
AFIT/GE/EE/83D-36
4& NAME OF PERFORMING ORGANIZATION b. OFFICE SYMBOL 7&. NAME OF MONITORING ORGANIZATION( i fappUcabiGe)
School of Engineering AFIT/ENGSe. ADDRESS (City. Stte and ZIP Code) 7b. ADORESS (City. Stud ad ZIP Code)
Air Force Institute of Technology
Wright-Patterson AFB, Ohio 45433Ift NAME OF FUNDING/SPONSORING Sb. OFFICE SYMBOL 9. PROCUREMENT INSTRUMENT IDENTIFICATION NUMBER
ORGANIZATION (If apflcable)
Sc. ADDRESS (City. State end ZIP Coce) 10. SOURCE OF FUNDING NOS.
PROGRAM PROJECT TASK WORK UNITELEMENT NO. NO. NO. NO.
11. TITLE (Include Selwiuty C1ausErflcauon)
See Box 1912. PERSONAL AUTHOR(S)
nsur F. Krilic, Lt Col. Yugoslavian Air ForceTYPE OF REPORT 13b. TIME COVERED 14. DATE OF REPORT (Yr.. Mo.. Day' 15. PAGE COUNT
. I FROM _Th TO 1983 Deember 2a1.8Is. SUPPLEMENTARY NOTATION aowed tox 0,€ ZIC h03 lW 1A' IN0-Iv
17. COSATI CODES 1. SUBJECT TERMS (Continue on reverse if ." IVJ ff!O& TkVWJw)
PIEL O GROUP SUBsR. Avionics, Digital Avionics, Computer Networks,Network Architecture
13. AMTRACT (Contlnue on uwerue if necemaiy and IdentIfy by block number)
Title: A STUDY OF OPTIMAL COMPUTER NETWORK ARCHITECTUREFOR DIGITAL AVIONICS SYSTEMS
Thesis Chairman: W. D. Seward, Ph.D., Major, USAF
:1 " OISTRIBUTION/AVAILASILITY OF ABSTRACT 21. ABSTRACT SECURITY CLASSIFICATION
CLASSIPIEO/UNLII MITE O SAME AS RPT. C1 O TIC USERS 03 UNCLASSIFIED
. 22a NAME OF RESPONSIBLE INDIVIDUAL 22b. TELEPHONE NUMBER 22c. OFFICE SYMBOL
00 FORM 1473, 83 APR EDITION OF I JAN 73 IS OBSOLETE. TINCT.ARSTFTDSECURITY CLASSIFICATION OF THIS PA
In this paper, current theoretical considerations inavailable literature have been used to sort out the essen-tial figures of merit of computer network architectures fordigital avionic systems. Fourteen different approaches tothe same problem of data multiplexing in avionics systemsare analyzed according to the key issues. Conclusions drawnare used to define the *optimal* computer network architec-ture for digital avionics.
The Self-Managing Multiplex System (SMS) is conceptu-ally designed with respect to the "optimal* characteristics,along with the discussions of some trade-offs that had to bemade. The burst errors self-correcting feature of the broad-cast-acknowledgments in the SMS concept seems to deservesome sort of testing in practice. It is recommended that adetailed simulation study should be performed later and ahot bench built up using the latest technologies that exist.
-A
-1
9.4